Network Working Group T. Connolly
Request for Comments: 1693 P. Amer
Category: EXPerimental P. Conrad
University of Delaware
November 1994
An Extension to TCP : Partial Order Service
Status of This Memo
This memo defines an Experimental Protocol for the Internet
community. This memo does not specify an Internet standard of any
kind. Discussion and suggestions for improvement are requested.
Distribution of this memo is unlimited
IESG Note:
Note that the work contained in this memo does not describe an
Internet standard. The Transport AD and Transport Directorate do not
recommend the implementation of the TCP modifications described.
However, outside the context of TCP, we find that the memo offers a
useful analysis of how misordered and incomplete data may be handled.
See, for example, the discussion of Application Layer Framing by D.
Clark and D. Tennenhouse in, "Architectural Considerations for a New
Generation of Protocols", SIGCOM 90 Proceedings, ACM, September 1990.
Abstract
This RFCintrodUCes a new transport mechanism for TCP based upon
partial ordering. The aim is to present the concepts of partial
ordering and promote discussions on its usefulness in network
communications. Distribution of this memo is unlimited.
Introduction
A service which allows partial order delivery and partial reliability
is one which requires some, but not all objects to be received in the
order transmitted while also allowing objects to be transmitted
unreliably (i.e., some may be lost).
The realization of such a service requires, (1) communication and/or
negotiation of what constitutes a valid ordering and/or loss-level,
and (2) an algorithm which enables the receiver to ascertain the
deliverability of objects as they arrive. These issues are addressed
here - both conceptually and formally - summarizing the results of
research and initial implementation efforts.
The authors envision the use of a partial order service within a
connection-oriented, transport protocol such as TCP providing a
further level of granularity to the transport user in terms of the
type and quality of offered service. This RFCfocuses specifically
on extending TCP to provide partial order connections.
The idea of a partial order service is not limited to TCP. It may be
considered a useful option for any transport protocol and we
encourage researchers and practitioners to investigate further the
most effective uses for partial ordering whether in a next-generation
TCP, or another general purpose protocol such as XTP, or perhaps
within a "special purpose" protocol tailored to a specific
application and network profile.
Finally, while the crux of this RFCrelates to and introduces a new
way of considering object ordering, a number of other classic
transport mechanisms are also seen in a new light - among these are
reliability, window management and data acknowledgments.
KeyWords: partial order, quality of service, reliability, multimedia,
client/server database, Windows, transport protocol
Table of Contents
1. Introduction and motivation .................................. 3
2. Partial Order Delivery ....................................... 4
2.1 Example 1: Remote Database .................................. 4
2.2 Example 2: Multimedia ....................................... 8
2.3 Example 3: Windows Screen Refresh ........................... 9
2.4 Potential Savings ........................................... 10
3. Reliability vs. Order ........................................ 12
3.1 Reliability Classes ......................................... 13
4. Partial Order Connection ..................................... 15
4.1 Connection Establishment .................................... 16
4.2 Data Transmission ........................................... 19
4.2.1 Sender .................................................... 22
4.2.2 Receiver .................................................. 25
5. Quantifying and Comparing Partial Order Services ............. 30
6. Future Direction ............................................. 31
7. Summary ...................................................... 32
8. References ................................................... 34
Security Considerations ......................................... 35
Authors' Addresses .............................................. 36
1. Introduction and motivation
Current applications that need to communicate objects (i.e., octets,
packets, frames, protocol data units) usually choose between a fully
ordered service such as that currently provided by TCP and one that
does not guarantee any ordering such as that provided by UDP. A
similar "all-or-nothing" choice is made for object reliability -
reliable connections which guarantee all objects will be delivered
verses unreliable data transport which makes no guarantee. What is
more appropriate for some applications is a partial order and/or
partial reliability service where a subset of objects being
communicated must arrive in the order transmitted, yet some objects
may arrive in a different order, and some (well specified subset) of
the objects may not arrive at all.
One motivating application for a partial order service is the
emerging area of multimedia communications. Multimedia traffic is
often characterized either by periodic, synchronized parallel streams
of information (e.g., combined audio-video), or by structured image
streams (e.g., displays of multiple overlapping and nonoverlapping
windows). These applications have a high degree of tolerance for
less-than-fully-ordered data transport as well as data loss. Thus
they are ideal candidates for using a partial order, partial
reliability service. In general, any application which communicates
parallel and/or independent data structures may potentially be able
to profit from a partial order service.
A second application that could benefit from a partial order service
involves remote or distributed databases. Imagine the case where a
database user transmitting queries to a remote server expects objects
(or records) to be returned in some order, although not necessarily
total order. For example a user writing an SQL data query might
specify this with the "order by" clause. There exist today a great
number of commercial implementations of distributed databases which
utilize - and thus are penalized by - an ordered delivery service.
Currently these applications must use and pay for a fully
ordered/fully reliable service even though they do not need it. The
introduction of partial services allows applications to lower the
demanded quality of service (QOS) of the communication assuming that
such a service is more efficient and less costly. In effect, a
partial order extends the service level from two extremes - ordered
and unordered - to a range of discreet values encompassing both of
the extremes and all possible partial orderings in between. A
similar phenomenon is demonstrated in the area of reliability.
It is worth mentioning that a TCP implementation providing a partial
order service, as described here, would be able to communicate with a
non-partial order implementation simply by recognizing this fact at
connection establishment - hence this extension is backward
compatible with earlier versions of TCP. Furthermore, it is
conceivable for a host to support the sending-half (or receiving-
half) of a partial order connection alone to reduce the size of the
TCP as well as the effort involved in the implementation. Similar
"levels of conformance" have been proposed in other internet
extensions such as [Dee89] involving IP multicasting.
This RFCproceeds as follows. The principles of partial order
delivery, published in [ACCD93a], are presented in Section 2. The
notion of partial reliability, published in [ACCD93b], is introduced
in Section 3 followed by an explanation of "reliability classes".
Then, the practical issues involved with setting up and maintaining a
Partial Order Connection (POC) within a TCP framework are addressed
in Section 4 looking first at connection establishment, and then
discussing the sender's role and the receiver's role. Section 5
provides insights into the expected performance improvements of a
partial order service over an ordered service and Section 6 discusses
some future directions. Concluding remarks are given in Section 7.
2. Partial Order Delivery
Partial order services are needed and can be employed as soon as a
complete ordering is not mandatory. When two objects can be
delivered in either order, there is no need to use an ordered service
that must delay delivery of the second one transmitted until the
first arrives as the following examples demonstrate.
2.1 Example 1: Remote Database
Simpson's Sporting Goods (SSG) has recently installed a state-of-
the-art enterprise-wide network. Their first "network application"
is a client/server SQL database with the following four records,
numbered {1 2 3 4} for convenience:
SALESPERSON LOCATION CHARGES DESCRIPTION
------------- ----------------- --------- -----------------
1 Anderson Atlanta, GA $4,200 Camping Gear
2 Baker Boston, MA $849 Camping Gear
3 Crowell Boston, MA $9,500 Sportswear
4 Dykstra Wash., DC $1,000 Sportswear
SSG employees running the client-side of the application can query
the database server from any location in the enterprise net using
standard SQL commands and the results will be displayed on their
screen. From the employee's perspective, the network is completely
reliable and delivers data (records) in an order that conforms to
their SQL request. In reality though, it is the transport layer
protocol which provides the reliability and order on top of an
unreliable network layer - one which introduces loss, duplication,
and disorder.
Consider the four cases in Figure 1 - in the first query (1.a),
ordered by SALESPERSON, the records have only one acceptable order at
the destination, 1,2,3,4. This is evident due to the fact that there
are four distinct salespersons. If record 2 is received before
record 1 due to a network loss during transmission, the transport
service can not deliver it and must therefore buffer it until record
1 arrives. An ordered service, also referred to as a virtual circuit
or FIFO channel, provides the desired level of service in this case.
At the other extreme, an unordered service is motivated in Figure 1.d
where the employee has implicitly specified that any ordering is
valid simply by omitting the "order by" clause. Here any of 4! = 24
delivery orderings would satisfy the application, or from the
transport layer's point of view, all records are immediately
deliverable as soon as they arrive from the network. No record needs
to buffered should it arrive out of sequential order. As notation, 4
ordered objects are written 1;2;3;4 and 4 unordered objects are
written using a parallel operator: 1234.
Figures 1.b and 1.c demonstrate two possible partial orders that
permit 2 and 4 orderings respectively at the destination. Using the
notation just described, the valid orderings for the query in 1.b are
specified as 1;(23);4, which is to say that record 1 must be
delivered first followed by record 2 and 3 in either order followed
by record 4. Likewise, the ordering for 1.c is (12);(34). In
these two cases, an ordered service is too strict and an unordered
service is not strict enough.
+-----------------------------------------------------------------+
SELECT SALESPERSON, LOCATION, CHARGES, DESCRIPTION
FROM BILLING_TABLE
SALESPERSON LOCATION CHARGES DESCRIPTION
------------- ----------------- --------- ---------------
1 Anderson Atlanta, GA $4,200 Camping Gear
2 Baker Boston, MA $849 Camping Gear
3 Crowell Boston, MA $9,500 Sportswear
4 Dykstra Wash., DC $1,000 Sportswear
+=================================================================+
a - ORDER BY SALESPERSON
1,2,3,4 1,2,3,4
Sender -----------> NETWORK --------------> Receiver
(1 valid ordering)
+-----------------------------------------------------------------+
b - ORDER BY LOCATION
1,2,3,4
1,2,3,4 1,3,2,4
Sender -----------> NETWORK --------------> Receiver
(2 valid orderings)
+-----------------------------------------------------------------+
c - ORDER BY DESCRIPTION
1,2,3,4
2,1,3,4
1,2,3,4 1,2,4,3
2,1,4,3
Sender -----------> NETWORK --------------> Receiver
(4 valid orderings)
+-----------------------------------------------------------------+
d - (no order by clause)
1,2,3,4
1,2,4,3
1,2,3,4 ...
4,3,2,1
Sender -----------> NETWORK --------------> Receiver
(4!=24 valid orderings)
+-----------------------------------------------------------------+
Figure 1: Ordered vs. Partial Ordered vs. Unordered Delivery
It is vital for the transport layer to recognize the exact
requirements of the application and to ensure that these are met.
However, there is no inherent need to exceed these requirements; on
the contrary, by exceeding these requirements unecessary resources
are consumed. This example application requires a reliable
connection - all records must eventually be delivered - but has some
flexibility when it comes to record ordering.
In this example, each query has a different partial order. In total,
there exist 16 different partial orders for the desired 4 records.
For an arbitrary number of objects N, there exist many possible
partial orders each of which accepts some number of valid orderings
between 1 and N! (which correspond to the ordered and unordered
cases respectively). For some classes of partial orders, the number
of valid orderings can be calculated easily, for others this
calculation is intractable. An in-depth discussion on calculating
and comparing the number of orderings for a given partial order can
be found in [ACCD93a].
2.2 Example 2: Multimedia
A second example application that motivates a partial order service
is a multimedia broadcast involving video, audio and text components.
Consider an extended presentation of the evening news - extended to
include two distinct audio channels, a text suBTitle and a closed-
captioned sign language video for the hearing impaired, in addition
to the normal video signal, as modeled by the following diagram.
(left audio) (right audio)
+------+ +------+
++++ ++++
++++ ++++
+------+ +------+
===================================================
I +---------------+I
I I
I (hand signs) I
I I
I +---------------+I
I I
I I
I (Main Video) I
I I
I I
I I
I I
I +------------------------------------------+ I
I (text subtitle) I
I +------------------------------------------+ I
I I
===================================================
Figure 2: Multimedia broadcast example
The multimedia signals have differing characteristics. The main video
signal may consist of full image graphics at a rate of 30 images/sec
while the video of hand signs requires a lower quality, say 10
images/sec. Assume the audio signals are each divided into 60 sound
fragments/sec and the text object each second consists of either (1)
new text, (2) a command to keep the previous second of text, or (3) a
command for no subtitle.
During a one-second interval of the broadcast, a sender transmits 30
full-motion video images, 10 closed-captioned hand sign images, 60
packets of a digitized audio signal for each of the audio streams and
a single text packet. The following diagram then might represent the
characteristics of the multimedia presentation in terms of the media
types, the number of each, and their ordering. Objects connected by a
horizontal line must be received in order, while those in parallel
have no inherent ordering requirement.
+----------------------------------------------------------------------+
-o--o--o--o--o--o--o--o--o-...-o--o--o- right audio
(60/sec)
-o--o--o--o--o--o--o--o--o-...-o--o--o- left audio
(60/sec)
---o------o------o------o------...------o--- normal video
(30/sec)
-----------o-------------------o--...--------o-- hand signs
(10/sec)
-----------------------------o-----...----------- text
(1/sec)
+----------------------------------------------------------------------+
Figure 3: Object ordering in multimedia application
Of particular interest to our discussion of partial ordering is the
fact that, while objects of a given media type generally must be
received in order, there exists flexibility between the separate
"streams" of multimedia data (where a "stream" represents the
sequence of objects for a specific media type). Another significant
characteristic of this example is the repeating nature of the object
orderings. Figure 3 represents a single, one-second, partial order
snapshot in a stream of possibly thousands of repeating sequential
periods of communication.
It is assumed that further synchronization concerns in presenting the
objects are addressed by a service provided on top of the proposed
partial order service. Temporal ordering for synchronized playback
is considered, for example, in [AH91, HKN91].
2.3 Example 3: Windows Screen Refresh
A third example to motivate a partial order service involves
refreshing a workstation screen/display containing multiple windows
from a remote source. In this case, objects (icons, still or video
images) that do not overlap have a "parallel" relationship (i.e.,
their order of refreshing is independent) while overlapping screen
objects have a "sequential" relationship and should be delivered in
order. Therefore, the way in which the windows overlap induces a
partial order.
Consider the two cases in Figure 4. A sender wishes to refresh a
remote display that contains four active windows (objects) named {1 2
3 4}. Assume the windows are transmitted in numerical order and the
receiving application refreshes windows as soon as the transport
service delivers them. If the windows are configured as in Figure
4a, then there exist two different orderings for redisplay, namely
1,2,3,4 or 1,3,2,4. If window 2 is received before window 1, the
transport service cannot deliver it or an incorrect image will be
displayed. In Figure 4b, the structure of the windows results in six
possible orderings - 1,2,3,4 or 1,3,2,4 or 1,3,4,2 or 3,4,1,2 or
3,1,4,2 or 3,1,2,4.
+================================+============================+
a +-----------+ b +----------+
1 1
+----------+
+---------+ +----------+ +----- 2
2 ---- 3
+-----------+ +----------+
4 +----------+
+----- -------+ 3
+----------+
+-----------+ +------ 4
+----------+
1;(23);4 (1;2)(3;4)
+================================+============================+
Figure 4: Window screen refresh
2.4 Potential Savings
In each of these examples, the valid orderings are strictly dependent
upon, and must be specified by the application. Intuitively, as the
number of acceptable orderings increases, the amount of resources
utilized by a partial order transport service, in terms of buffers
and retransmissions, should decrease as compared to a fully ordered
transport service thus also decreasing the overall cost of the
connection. Just how much lower will depend largely upon the
flexibility of the application and the quality of the underlying
network.
As an indication of the potential for improved service, let us
briefly look at the case where a database has the following 14
records.
SALESPERSON LOCATION CHARGES DESCRIPTION
------------- ----------------- --------- ---------------
1 Anderson Washington $4,200 Camping Gear
2 Anderson PhilaDelphia $2,000 Golf Equipment
3 Anderson Boston $450 Bowling shoes
4 Baker Boston $849 Sportswear
5 Baker Washington $3,100 Weights
6 Baker Washington $2000 Camping Gear
7 Baker Atlanta $290 Baseball Gloves
8 Baker Boston $1,500 Sportswear
9 Crowell Boston $9,500 Camping Gear
10 Crowell Philadelphia $6,000 Exercise Bikes
11 Crowell New York $1,500 Sportswear
12 Dykstra Atlanta $1,000 Sportswear
13 Dykstra Dallas $15,000 Rodeo Gear
14 Dykstra Miami $3,200 Golf Equipment
Using formulas derived in [ACCD93a] one may calculate the total
number of valid orderings for any partial order that can be
represented in the notation mentioned previously. For the case where
a user specifies "ORDER BY SALESPERSON", the partial order above can
be expressed as,
(123);(45678);(91011);(121314)
Of the 14!=87,178,291,200 total possible combinations, there exist
25,920 valid orderings at the destination. A service that may
deliver the records in any of these 25,920 orderings has a great deal
more flexibility than in the ordered case where there is only 1 valid
order for 14 objects. It is interesting to consider the real
possibility of hundreds or even thousands of objects and the
potential savings in communication costs.
In all cases, the underlying network is assumed to be unreliable and
may thus introduce loss, duplication, and disorder. It makes no
sense to put a partial order service on top of a reliable network.
While the exact amount of unreliability in a network may vary and is
not always well understood, initial experimental research indicates
that real world networks, for example the service provided by the
Internet's IP level, "yield high losses, duplicates and reorderings
of packets" [AS93,BCP93]. The authors plan to conduct further
experimentation into measuring Internet network unreliability. This
information would say a great deal about the practical merit of a
partial order service.
3. Reliability vs. Order
While TCP avoids the loss of even a single object, in fact for many
applications, there exists a genuine ability to tolerate loss.
Losing one frame per second in a 30 frame per second video or losing
a segment of its accompanying audio channel is usually not a problem.
Bearing this in mind, it is of value to consider a quality of service
that combines a partial order with a level of tolerated loss (partial
reliability). Traditionally there exist 4 services: reliable-
ordered, reliable-unordered, unreliable-ordered, and unreliable-
unordered. See Figure 5. Reliable-ordered service (denoted by a
single point) represents the case where all objects are delivered in
the order transmitted. File transfer is an example application
requiring such a service.
reliable-ordered reliable-unordered
v v
zero loss-->*---------------------------------*
min loss--><-- <--
.
. <-- <--
<-- unreliable- <-- unreliable-
RELIABILITY ordered unordered
<-- <--
<-- <--
max loss-->
+-+--+--+--+--+--+--+--+--+--+--+-+
ordered partial ordered unordered
ORDER
Figure 5: Quality Of Service: Reliability vs. Order -
Traditional Service Types
In a reliable-unordered service (also a single point), all objects
must be delivered, but not necessarily according to the order
transmitted; in fact, any order will suffice. Some transaction
processing applications such as credit card verification require such
a service.
Unreliable-ordered service allows some objects to be lost. Those
that are delivered, however, must arrive in relative order (An
"unreliable" service does not necessarily lose objects; rather, it
may do so without failing to provide its advertised quality of
service; e.g., the postal system provides an unreliable service).
Since there are varying degrees of unreliability, this service is
represented by a set of points in Figure 5. An unreliable-ordered
service is applicable to packet-voice or teleconferencing
applications.
Finally unreliable-unordered service allows objects to be lost and
delivered in any order. This is the kind of service used for normal
e-mail (without acknowledgment receipts) and electronic announcements
or junk e-mail.
As mentioned previously, the concept of a partial order expands the
order dimension from the two extremes of ordered and unordered to a
range of discrete possibilities as depicted in Figure 6.
Additionally, as will be discussed presently, the notion of
reliability is extended to allow for varying degrees of reliability
on a per-object basis providing even greater flexibility and improved
resource utilization.
reliable-PO
v v v v v v v v v v v v
zero loss-->*---------------------------------*
min loss--> . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . .
RELIABILITY . . . unreliable-PO . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
max loss--> . . . . . . . . . . .
+-+--+--+--+--+--+--+--+--+--+--+-+
ordered partial ordered unordered
ORDER
Figure 6: Quality Of Service: Reliability vs. Order - Partial
Order Service
3.1 Reliability Classes
When considering unreliable service, one cannot assume that all
objects are equal with regards to their reliability. This
classification is reasonable if all objects are identical (e.g.,
video frames in a 30 frame/second film). Many applications, such as
multimedia systems, however, often contain a variety of object types.
Thus three object reliability classes are proposed: BART-NL, BART-L,
and NBART-L. Objects are assigned to one of these classes depending
on their temporal value as will be show presently.
BART-NL objects must be delivered to the destination. These objects
have temporal value that lasts for an entire established connection
and require reliable delivery (NL = No Loss allowed). An example of
BART-NL objects would be the database records in Example 2.1 or the
windows in the screen refresh in Example 2.3. If all objects are of
type BART-NL, the service is reliable. One possible way to assure
eventual delivery of a BART-NL object in a protocol is for the sender
to buffer it, start a timeout timer, and retransmit it if no ACK
arrives before the timeout. The receiver in turn returns an ACK when
the object has safely arrived and been delivered (BART = Buffers,
ACKs, Retransmissions, Timers).
BART-L objects are those that have temporal value over some
intermediate amount of time - enough to permit timeout and
retransmission, but not everlasting. Once the temporal value of
these objects has expired, it is better to presume them lost than to
delay further the delivery pipeline of information. One possibility
for deciding when an object's usefulness has expired is to require
each object to contain information defining its precise temporal
value [DS93]. An example of a BART-L object would be a movie
subtitle, sent in parallel with associated film images, which is
valuable any time during a twenty second film sequence. If not
delivered sometime during the first ten seconds, the subtitle loses
its value and can be presumed lost. These objects are buffered-
ACKed-retransmitted up to a certain point in time and then presumed
lost.
NBART-L objects are those with temporal values too short to bother
timing out and retransmitting. An example of a NBART-L object would
be a single packet of speech in a packetized phone conversation or
one image in a 30 image/sec film. A sender transmits these objects
once and the service makes a best effort to deliver them. If the one
attempt is unsuccessful, no further attempts are made.
An obvious question comes to mind - what about NBART-NL objects? Do
such objects exist? The authors have considered the notion of
communicating an object without the use of BART and still being able
to provide a service without loss. Perhaps with the use of forward
error correction this may become a viable alternative and could
certainly be included in the protocol. However, for our purposes in
this document, only the first three classifications will be
considered.
While classic transport protocols generally treat all objects
equally, the sending and receiving functions of a protocol providing
partial order/partial reliability service will behave differently for
each class of object. For example, a sender buffers and, if
necessary, retransmits any BART-NL or BART-L objects that are not
acknowledged within a predefined timeout period. On the contrary,
NBART-L objects are forgotten as soon as they are transmitted.
4. Partial Order Connection
The implementation of a protocol that provides partial order service
requires, at a minimum, (1) communication of the partial ordering
between the two endpoints, and (2) dynamic evaluation of the
deliverability of objects as they arrive at the receiver. In
addition, this RFCdescribes the mechanisms needed to (3) initiate a
connection, (4) provide varying degrees of reliability for the
objects being transmitted, and (5) improve buffer utilization at the
sender based on object reliability.
Throughout the discussion of these issues, the authors use the
generic notion of "objects" in describing the service details. Thus,
one of the underlying requirements of a partial order service is the
ability to handle such an abstraction (e.g., recognize object
boundaries). The details of object management are implementation
dependent and thus are not specified in this RFC. However, as this
represents a potential fundamental change to the TCP protocol, some
discussion is in order.
At one extreme, it is possible to consider octets as objects and
require that the application specify the partial order accordingly
(octet by octet). This likely would entail an inordinate amount of
overhead, processing each octet on an individual basis (literally
breaking up contiguous segments to determine which, if any, octets
are deliverable and which are not). At the other extreme, the
transport protocol could maintain object atomicity regardless of size
- passing arbitrarily large data structures to IP for transmission.
At the sending side of the connection this would actually work since
IP is prepared to perform source fragmentation, however, there is no
guarantee that the receiving IP will be able to reassemble the
fragments! IP relies on the TCP max segment size to prevent this
situation from occurring[LMKQ89].
A more realistic approach given the existing IP constraints might be
to maintain the current notion of a TCP max segment size for the
lower-layer interface with IP while allowing a much larger object
size at the upper-layer interface. Of course this presents some
additional complexities. First of all, the transport layer will now
have to be concerned with fragmentation/reassembly of objects larger
than the max segment size and secondly, the increased object sizes
will require significantly more buffer space at the receiver if we
want to buffer the object until it arrives in entirety.
Alternatively, one may consider delivering "fragments" of an object
as they arrive as long as the ordering of the fragments is correct
and the application is able to process the fragments (this notion of
fragmented delivery is discussed further in Section 6).
4.1 Connection Establishment
By extending the transport paradigm to allow partial ordering and
reliability classes, a user application may be able to take advantage
of a more efficient data transport facility by negotiating the
optimal service level which is required - no more, no less. This is
accomplished by specifying these variables as QOS parameters or, in
TCP terminology, as options to be included in the TCP header [Pos81].
A TCP implementation that provides a partial order service requires
the use of two new TCP options. The first is an enabling option
"POC-permitted" (Partial Order Connection Permitted) that may be used
in a SYN segment to request a partial order service. The other is
the "POC-service-profile" option which is used periodically to
communicate the service characteristics. This second option may be
sent only after successful transmission and acknowledgment of the
POC-permitted option.
A user process issuing either an active or passive OPEN may choose to
include the POC-permitted option if the application can benefit from
the use of a partial order service and in fact, in cases where the
viability of such service is unknown, it is suggested that the option
be used and that the decision be left to the user's peer.
For example, a multimedia server might issue a passive <SYN> with the
POC-permitted option in preparation for the connection by a remote
user.
Upon reception of a <SYN> segment with the POC-permitted option, the
receiving user has the option to respond with a similar POC-permitted
indication or may reject a partial order connection if the
application does not warrant the service or the receiving user is
simply unable to provide such a service (e.g., does not recognize the
POC-permitted option).
In the event that simultaneous initial <SYN> segments are exchanged,
the TCP will initiate a partial order connection only if both sides
include the POC-permitted option.
A brief example should help to demonstrate this procedure. The
following notation (a slight simplification on that employed in RFC
793) will be used. Each line is numbered for reference purposes.
TCP-A (on the left) will play the role of the receiver and TCP-B will
be the sender. Right arrows (-->) indicate departure of a TCP
segment from TCP-A to TCP-B, or arrival of a segment at B from A.
Left arrows indicate the reverse. TCP states represent the state
AFTER the departure or arrival of the segment (whose contents are
shown in the center of the line). Liberties are taken with the
contents of the segments where only the fields of interest are shown.
TCP-A TCP-B
1. CLOSED LISTEN
2. SYN-SENT --> <CTL=SYN><POC-perm> --> SYN-RECEIVED
3. ESTABLISHED <-- <CTL=SYN,ACK><POC-perm> <-- SYN-RECEIVED
4. ESTABLISHED --> <CTL=ACK> --> ESTABLISHED
Figure 7. Basic 3-Way handshake for a partial order connection
In line 1 of Figure 7, the sending user has already issued a passive
OPEN with the POC-permitted option and is waiting for a connection.
In line 2, the receiving user issues an active OPEN with the same
option which in turn prompts TCP-A to send a SYN segment with the
POC-permitted option and enter the SYN-SENT state. TCP-B is able to
confirm the use of a PO connection and does so in line 3, after which
TCP-A enters the established state and completes the connection with
an ACK segment in line 4.
In the event that either side is unable to provide partial order
service, the POC-permitted option will be omitted and normal TCP
processing will ensue.
For completeness, the authors include the following specification for
both the POC-permitted option and the POC-service-profile option in a
format consistent with the TCP specification document [Pos81].
TCP POC-permitted Option:
Kind: 9 Length: - 2 bytes
+-----------+-------------+
Kind=9 Length=2
+-----------+-------------+
TCP POC-service-profile Option:
Kind: 10 Length: 3 bytes
1 bit 1 bit 6 bits
+----------+----------+------------+----------+--------+
Kind=10 Length=3 Start_flag End_flag Filler
+----------+----------+------------+----------+--------+
The first option represents a simple indicator communicated between
the two peer transport entities and needs no further explanation.
The second option serves to communicate the information necessary to
carry out the job of the protocol - the type of information which is
typically found in the header of a TCP segment - and raises some
interesting questions.
Standard TCP maintains a 60-byte maximum header size on all segments.
The obvious intuition behind this rule is that one would like to
minimize the amount of overhead information present in each packet
while simultaneously increasing the payload, or data, section. While
this is acceptable for most TCP connections today, a partial-order
service would necessarily require that significantly more control
information be passed between transport entities at certain points
during a connection. Maintaining the strict interpretation of this
rule would prove to be inefficient. If, for example, the service
profile occupied a total of 400 bytes (a modest amount as will be
confirmed in the next section), then one would have to fragment this
information across at least 10 segments, allocating 20 bytes per
segment for the normal TCP header.
Instead, the authors propose that the service profile be carried in
the data section of the segment and that the 3-byte POC-service-
profile option described above be placed in the header to indicate
the presence of this information. Upon reception of such a segment,
the TCP extracts the service profile and uses it appropriately as
will be discussed in the following sections.
The option itself, as shown here, contains two 1-bit flags necessary
to handle the case where the service profile does not fit in a single
TCP segment. The "Start_flag" indicates that the information in the
data section represents the beginning of the service profile and the
"End_flag" represents the converse. For service profiles which fit
completely in a single segment, both flags will be set to 1.
Otherwise, the Start_flag is set in the initial segment and the
End_flag in the final segment allowing the peer entity to reconstrcut
the entire service profile (using the normal sequence numbers in the
segment header). The "Filler" field serves merely to complete the
third byte of the option.
Note that the length of the service profile may vary during the
connection as the order or reliability requirements of the user
change but this length must not exceed the buffering ability of the
peer TCP entity since the entire profile must be stored. The exact
makeup of this data structure is presented in Section 4.2.
4.2 Data Transmission
Examining the characteristics of a partial order TCP in chronological
fashion, one would start off with the establishment of a connection
as described in Section 4.1. After which, although both ends have
acknowledged the acceptability of partial order transport, neither
has actually begun a partial order transmission - in other words,
both the sending-side and the receiving-side are operating in a
normal, ordered-reliable mode. For the subsequent discussion, an
important distinction is made in the terms sending-side and
receiving-side which refer to the data flow from the sender and that
from the receiver, respectively.
For the partial ordering to commence, the TCP must be made aware of
the acceptable object orderings and reliability for both the send-
side and receive-side of the connection for a given set of objects
(hereafter referred to as a "period"). This information is contained
in the service profile and it is the responsibility of the user
application to define this profile. Unlike standard TCP where
applications implicitly define a reliable, ordered profile; with
partial order TCP, the application must explicity define a profile.
The representation of the service profile is one of the concerns for
the transport protocol. It would be useful if the TCP could encode a
partial ordering in as few bits as possible since these bits will be
transmitted to the destination each time the partial order changes.
A matrix representation appears to be well-suited to encoding the
partial order and a vector has been proposed to communicate and
manage the reliability ASPects of the service. Temporal values may
be included within the objects themselves or may be defined as a
function of the state of the connection [DS93]. Using these data
structures, the complete service profile would include (1) a partial
order matrix, (2) a reliability vector and (3) an object_sizes vector
which represents the size of the objects in octets (see
[ACCD93a,CAC93] for a discussion on alternative structures for these
variables).
Throughout this section, we use the following service profile as a
running example. Shown here is a partial order matrix and graphical
representation for a simple partial order with 6 objects -
((1;2)(3;4)5);6. In the graphical diagram, arrows (-->) denote
sequential order and objects in parallel can be delivered in either
order. So in this example, object 2 must be delivered after object
1, object 4 must be delivered after object 3, and object 6 must be
delivered after objects 1 through 5 have all been delivered. Among
the 6 objects, there are 30 valid orderings for this partial order
(each valid ordering is known as a linear extension of the partial
order).
1 2 3 4 5 6
+-------------+
1 - 1 0 0 0 1
2 - - 0 0 0 1 -->1-->-->2-->
3 - - - 1 0 1
4 - - - - 0 1 -->3-->-->4-->-->6-->
5 - - - - - 1
6 - - - - - - ------>5------>
+-------------+
PO Matrix PO Graph
In the matrix, a 1 in row i of column j denotes that object i must be
delivered before object j. Note that if objects are numbered in any
way such that 1,2,3,...,N is a valid ordering, only the upper right
triangle of the transitively closed matrix is needed [ACCD93a].
Thus, for N objects, the partial order can be encoded in (N*(N-1)/2)
bits.
The reliability vector for the case where reliability classes are
enumerated types such as {BART-NL=1, BART-L=2, NBART-L = 3} and all
objects are BART-NL would simply be, <1, 1, 1, 1, 1, 1>. Together
with the object_sizes vector, the complete service profile is
described.
This information must be packaged and communicated to the sending TCP
before the first object is transmitted using a TCP service primitive
or comparable means depending upon the User/TCP interface. Once the
service profile has been specified to the TCP, it remains in effect
until the connection is closed or the sending user specifies a new
service profile. In the event that the largest object size can not
be processed by the receiving TCP, the user application is informed
that the connection cannot be maintained and the normal connection
close procedure is followed.
Typically, as has been described here, the service profile definition
and specification is handled at the sending end of the connection,
but there could be applications (such as the screen refresh) where
the receiving user has this knowledge. Under these circumstances the
receiving user is obliged to transmit the object ordering on the
return side of the connection (e.g., when making the request for a
screen refresh) and have the sender interpret this data to be used on
the send side of the connection.
Requiring that the sending application specify the service profile is
not an arbitrary choice. To ensure proper object identification, the
receiving application must transmit the new object numbering to the
sending application (not the sending transport layer). Since the
sending application must receive this information in any case, it
simplifies matters greatly to require that the sending application be
the only side that may specify the service profile to the transport
layer.
Consider now the layered architecture diagram in Figure 8 and assume
that a connection already is established. Let us now say that UserA
specifies the service profile for the sending-side of the connection
via its interface with TCP-A. TCP-A places the profile in the header
of one or more data packets (depending upon the size of the service
profile, the profile may require several packets), sets the POC-
service-profile option and passes it to IP for transmission over the
network. This packet must be transmitted reliably, therefore TCP-A
buffers it and starts a normal retransmit timer. Subsequently, the
service profile arrives at the destination node and is handed to
TCP-B (as indicated by the arrows in Figure 8). TCP-B returns an
acknowledgment and immediately adopts the service profile for one
direction of data flow over the connection. When the acknowledgment
arrives back at TCP-A, the cycle is complete and both sides are now
able to use the partial order service.
+--------+ +----------+
Service UserA UserB
Profile +--------+ +----------+
v
+---------+ +-----------+ Service
TCP-A TCP-B Profile
+---------+ +-----------+ ^
+---------------------------------------+
v
------> ---- Service Profile -------------> ----->
+---------------------------------------+
Figure 8. Layered Communication Architecture
Note that one of the TCP entities learns of the profile via its user
interface, while the other TCP entity is informed via its network
interface.
For the remaining discussions, we will assume that a partial order
profile has been successfully negotiated for a single direction of
the connection (as depicted in Figure 8) and that we may now speak of
a "sending TCP" (TCP-A) and a "receiving TCP" (TCP-B). As such,
TCP-A refers to the partial order data stream as the "send-side" of
the connection, while TCP-B refers to the same data stream as the
"receive-side".
Having established a partial order connection, the communicating TCPs
each have their respective jobs to perform to ensure proper data
delivery. The sending TCP ascertains the object ordering and
reliability from the service profile and uses this information in its
buffering/retransmission policy. The receiver modifications are more
significant, particularly the issues of object deliverability and
reliability. And both sides will need to redefine the notion of
window management. Let us look specifically at how each side of the
TCP connection is managed under this new paradigm.
4.2.1 Sender
The sender's concerns are still essentially four-fold - transmitting
data, managing buffer space, processing acknowledgments and
retransmitting after a time-out - however, each takes on a new
meaning in a partial order service. Additionally, the management of
the service profile represents a fifth duty not previously needed.
Taking a rather simplistic view, normal TCP output processing
involves (1) setting up the header, (2) copying user data into the
outgoing segment, (3) sending the segment, (4) making a copy in a
send buffer for retransmission and (5) starting a retransmission
timer. The only difference with a partial order service is that the
reliability vector must be examined to determine whether or not to
buffer the object and start a timer - if the object is classified as
NBART-L, then steps 4 and 5 are omitted.
Buffer management at the sending end of a partial order connection is
dependent upon the object reliability class and the object size.
When transmitting NBART-L objects the sender need not store the data
for later possible retransmission since NBART-L objects are never
retransmitted. The details of buffer management - such as whether to
allocate fixed-size pools of memory, or perhaps utilize a dynamic
heap allocation strategy - are left to the particular system
implementer.
Acknowledgment processing remains essentially intact -
acknowledgments are cumulative and specify the peer TCP's window
advertisement. However, determination of this advertisement is no
longer a trivial process dependent only upon the available buffer
space (this is discussed further in Section 4.2.2). Moreover, it
should be noted that the introduction of partial ordering and partial
reliability presents several new and interesting alternatives for the
acknowledgment policy. The authors are investigating several of
these strategies through a simulation model and have included a brief
discussion of these issues in Section 6.
The retransmit function of the TCP is entirely unchanged and is
therefore not discussed further.
For some applications, it may be possible to maintain the same
partial order for multiple periods (e.g., the application repeats the
same partial order). In the general case, however, the protocol must
be able to change the service profile during an existing connection.
When a change in the service profile is requested, the sending TCP is
obliged to complete the processing of the current partial order
before commencing with a new one. This ensures consistency between
the user applications in the event of a connection failure and
simplifies the protocol (future study is planned to investigate the
performance improvement gained by allowing concurrent different
partial orders). The current partial order is complete when all
sending buffers are free. Then negotiation of the new service
profile is performed in the same manner as with the initial profile.
Combining these issues, we propose the following simplified state
machine for the protocol (connection establishment and tear down
remains the same and is not show here).
(1)Send Request (5)Ack Arrival
+------+ +-----------+
V
+----------+ (4) New PO Profile +----------+
+----> -----------------------> PO <-----+
ESTAB
(2) SETUP
Ack +----- <----------------------- <-----+
Arrival +----------+ (7)PO Setup Complete +----------+
^
+------+ +---------+
(3)Timeout (6)Timeout
Event (1) - User Makes a Data Send Request
=========
If Piggyback Timer is set then
cancel piggyback timer
Package and send the object (with ACK for receive-side)
If object type = (BART-L,BART-NL) then
Store the object and start a retransmit timer
If sending window is full then
Block Event (1) - allow no further send requests from user
Event (2) - ACK Arrives
=========
If ACKed object(s) is buffered then
Release the buffer(s) and stop the retransmit timer(s)
Extract the peer TCP's window advertisement
If remote TCP's window advertisement > sending window then
Enable Event (1)
If remote TCP's window advertisement <= sending window then
Block Event (1) - allow no further send requests from user
Adjust sending window based on received window advertisement
Event (3) - Retransmit Timer Expires
=========
If Piggyback Timer is set then
cancel piggyback timer
Re-transmit the segment (with ACK for receive-side)
Restart the timer
Event (4) - PO Service Profile Arrives at the User Interface
=========
Transition to the PO SETUP state
Store the Send-side PO service profile
Package the profile into 1 or more segments, setting the
POC-Service-Profile option on each
If Piggyback Timer is set then
cancel piggyback timer
Send the segment(s) (with ACK for receive-side)
Store the segment(s) and start a retransmit timer
Event (5) - ACK Arrival
=========
If ACKed object(s) is buffered then
Release the buffer(s) and stop the retransmit timer(s)
Extract the peer TCP's window advertisement
If all objects from previous service profile have been ACKed and
the new service profile has been ACKed then enable Event (7)
Event (6) - Retransmit Timer Expires
=========
If Piggyback Timer is set then
cancel piggyback timer
Re-transmit the segment (with ACK for receive-side)
Restart the timer
Event (7) - PO Setup Completed
=========
Transition to the ESTAB state and begin processing new service
profile
4.2.2 Receiver
The receiving TCP has additional decisions to make involving object
deliverability, reliability and window management. Additionally, the
service profile must be established (and re-established) periodically
and some special processing must be performed at the end of each
period.
When an object arrives, the question is no longer, "is this the next
deliverable object?", but rather, "is this ONE OF the next
deliverable objects?" Hence, it is convenient to think of a
"Deliverable Set" of objects with a partial order protocol. To
determine the elements of this set and answer the question of
deliverability, the receiver relies upon the partial order matrix
but, unlike the sender, the receiver dynamically updates the matrix
as objects are processed thus making other objects (possibly already
buffered objects) deliverable as well. A check of the object type
also must be performed since BART-NL and BART-L objects require an
ACK to be returned to the sender but NBART-L do not. Consider our
example from the previous section.
1 2 3 4 5 6
+-------------+
1 - 1 0 0 0 1
2 - - 0 0 0 1 -->1-->-->2-->
3 - - - 1 0 1
4 - - - - 0 1 -->3-->-->4-->-->6-->
5 - - - - - 1
6 - - - - - - ------>5------>
+-------------+
PO Matrix PO Graph
When object 5 arrives, the receiver scans column 5, finds that the
object is deliverable (since there are no 1's in the column) and
immediately delivers the object to the user application. Then, the
matrix is updated to remove the constraint of any object whose
delivery depends on object 5 by clearing all entries of row 5. This
may enable other objects to be delivered (for example, if object 2 is
buffered then the delivery of object 1 will make object 2
deliverable). This leads us to the next issue - delivery of stored
objects.
In general, whenever an object is delivered, the buffers must be
examined to see if any other stored object(s) becomes deliverable.
CAC93 describes an efficient algorithm to implement this processing
based on traversing the precedence graph.
Consideration of object reliability is interesting. The authors have
taken a polling approach wherein a procedure is executed
periodically, say once every 100 milliseconds, to evaluate the
temporal value of outstanding objects on which the destination is
waiting. Those whose temporal value has expired (i.e. which are no
longer useful as defined by the application) are "declared lost" and
treated in much the same manner as delivered objects - the matrix is
updated, and if the object type is BART-L, an ACK is sent. Any
objects from the current period which have not yet been delivered or
declared lost are candidates for the "Terminator" as the procedure is
called. The Terminator's criterion is not specifically addressed in
this RFC, but one example might be for the receiving user to
periodically pass a list of no-longer-useful objects to TCP-B.
Another question which arises is, "How does one calculate the send
and receive windows?" With a partial order service, these windows
are no longer contiguous intervals of objects but rather sets of
objects. In fact, there are three sets which are of interest to the
receiving TCP one of which has already been mentioned - the
Deliverable Set. Additionally, we can think of the Bufferable Set
and the Receivable Set. Some definitions are in order:
Deliverable Set: objects which can be immediately passed up to
the user.
Buffered Set: objects stored in a buffer awaiting delivery.
Bufferable Set: objects which can be stored but not immediately
delivered (due to some ordering constraint).
Receivable Set: union of the Deliverable Set and the Bufferable
Set (which are disjoint) - intuitively, all objects which
are "receivable" must be either "deliverable" or
"bufferable".
The following example will help to illustrate these sets. Consider
our simple service profile from earlier for the case where the size
of each object is 1 MByte and the receiver has only 2 MBytes of
buffer space (enough for 2 objects). Define a boolean vector of
length N (N = number of objects in a period) called the Processed
Vector which is used to indicate which objects from the current
period have been delivered or declared lost. Initially, all buffers
are empty and the PO Matrix and Processed Vector are as shown here,
1 2 3 4 5 6
+-------------+
1 - 1 0 0 0 1
2 - - 0 0 0 1
3 - - - 1 0 1
4 - - - - 0 1
5 - - - - - 1 [ F F F F F F ]
6 - - - - - - 1 2 3 4 5 6
+-------------+
PO Matrix Processed Vector
From the PO Matrix, it is clear that the Deliverable Set =
{(1,1),(1,3),(1,5)}, where (1,1) refers to object #1 from period #1,
asssuming that the current period is period #1.
The Bufferable Set, however, depends upon how one defines bufferable
objects. Several approaches are possible. The authors' initial
approach to determining the Bufferable Set can best be explained in
terms of the following rules,
Rule 1: Remaining space must be allocated for all objects from
period i before any object from period i+1 is buffered
Rule 2: In the event that there exists enough space to buffer
some but not all objects from a given period, space will
be reserved for the first objects (i.e. 1,2,3,...,k)
With these rules, the Bufferable Set = {(1,2),(1,4)}, the Buffered
Set is trivially equal to the empty set, { }, and the Receivable Set
= {(1,1),(1,2),(1,3),(1,4),(1,5)}.
Note that the current acknowledgment scheme uses the min and max
values in the Receivable Set for its window advertisement which is
transmitted in all ACK segments sent along the receive-side of the
connection (from receiver to sender). Moreover, the
"piggyback_delay" timer is still used to couple ACKs with return data
(as utilized in standard TCP).
Returning to our example, let us now assume that object 1 and then 3
arrive at the receiver and object 2 is lost. After processing both
objects, the PO Matrix and Processed Vector will have the following
updated structure,
1 2 3 4 5 6
+-------------+
1 - 0 0 0 0 0
2 - - 0 0 0 1
3 - - - 0 0 0
4 - - - - 0 1
5 - - - - - 1 [ T F T F F F ]
6 - - - - - - 1 2 3 4 5 6
+-------------+
PO Matrix Processed Vector
We can see that the Deliverable Set = {(1,2),(1,4),(1,5)}, but what
should the Bufferable Set consist of? Since only one buffer is
required for the current period's objects, we have 1 Mbyte of
additional space available for "future" objects and therefore include
the first object from period #2 in both the Bufferable and the
Receivable Set,
Deliverable Set = {(1,2),(1,4),(1,5)}
Bufferable Set = {(1,6),(2,1)}
Buffered Set = { }
Receivable Set = {(1,2),(1,4),(1,5),(1,6),(2,1)}
In general, the notion of window management takes on new meaning with
a partial order service. One may re-examine the classic window
relations with a partial order service in mind and devise new, less
restrictive relations which may shed further light on the operation
of such a service.
Two final details: (1) as with the sender, the receiver must
periodically establish or modify the PO service profile and (2) upon
processing the last object in a period, the receiver must re-set the
PO matrix and Processed vector to their initial states.
Let us look at the state machine and pseudo-code for the receiver.
(2)Data Segment Arrival (5)PO Profile fragment Arrival
+------+ +-------+
V (1)First PO Profile V
+---------+ fragment arrives +---------+(6) Data Segment
+----> -----------------------> <-----+ Arrival
ESTAB PO ------+
SETUP <-----+
(3) +----- <----------------------- ------+
Terminator+---------+ (9)PO Setup complete +---------+(7) Terminator
^ ^
+------+ +------+
(4)Piggyback Timeout (8)Piggyback Timeout
Event 1 - First PO Service Profile fragment arrives at network
======= interface
Transition to the PO SETUP state
Store the PO service profile (fragment)
Send an Acknowledgement of the PO service profile (fragment)
Event 2 - Data Segment Arrival
=======
If object is in Deliverable Set then
Deliver the object
Update PO Matrix and Processed Vector
Check buffers for newly deliverable objects
If all objects from current period have been processed then
Start the next period (re-initialize data structures)
Start piggyback_delay timer to send an ACK
Else if object is in Bufferable Set then
Store the object
Else
Discard object
Start piggyback_delay timer to send an ACK
Event 3 - Periodic call of the Terminator
=======
For all unprocessed objects in the current period do
If object is "no longer useful" then
Update PO Matrix and Processed Vector
If object is in a buffer then
Release the buffer
Check buffers for newly deliverable objects
If all objects from current period have been processed
then Start the next period (re-initialize data
structures)
Event 4 - Piggyback_delay Timer Expires
=======
Send an ACK
Disable piggyback_delay timer
Event 5 - PO Service Profile fragment arrives at network interface
=======
Store the PO service profile (fragment)
Send an Acknowledgement of the PO service profile (fragment)
If entire PO Service profile has been received then enable Event
(9)
Event 6 - Data Segment arrival
=======
(See event 2)
Event 7 - Periodic call of the terminator
=======
(See Event 3)
Event 8 - Piggyback_delay Timer Expires
=======
(See Event 4)
Event 9 - PO Setup Complete
=======
Transition to the ESTAB state
Note that, for reasons of clarity, we have used a transitively closed
matrix representation of the partial order. A more efficient
implementation based on an adjacency list representation of a
transitively reduced precedence graph results in a more efficient
running time [CAC93].
5. Quantifying and Comparing Partial Order Services
While ordered, reliable delivery is ideal, the existence of less-
than-ideal underlying networks can cause delays for applications that
need only partial order or partial reliability. By introducing a
partial order service, one may in effect relax the requirements on
order and reliability and presumably expect some savings in terms of
buffer utilization and bandwidth (due to fewer retransmissions) and
shorter overall delays. A practical question to be addressed is,
"what are the expected savings likely to be?"
As mentioned in Section 2, the extent of such savings will depend
largely on the quality of the underlying network - bandwidth, delay,
amount and distribution of loss/duplication/disorder - as well as the
flexibility of the partial order itself - specified by the PO matrix
and reliability vector. If the underlying network has no loss, a
partial order service essentially becomes an ordered service.
Collecting experimental data to ascertain realistic network
conditions is a straightforward task and will help to quantify in
general the value of a partial order service [Bol93]. But how can
one quantify and compare the cost of providing specific levels of
service?
Preliminary research indicates that the number of linear extensions
(orderings) of a partial order in the presence of loss effectively
measures the complexity of that order. The authors have derived
formulae for calculating the number of extensions when a partial
order is series-parallel and have proposed a metric for comparing
partial orders based on this number [ACCD93b]. This metric could be
used as a means for charging for the service, for example. What also
may be interesting is a specific head-to-head comparison between
different partial orders with varying degrees of flexibility. Work
is currently underway on a simulation model aimed at providing this
information. And finally, work is underway on an implementation of
TCP which includes partial order service.
6. Future Direction
In addition to the simulation and implementation work the authors are
pursuing several problems related to partial ordering which will be
mentioned briefly.
An interesting question arises when discussing the acknowledgment
strategy for a partial order service. For classic protocols, a
cumulative ACK of object i confirms all objects "up to and including"
i. But the meaning of "up to and including" with a partial order
service has different implications than with an ordered service.
Consider our example partial order, ((1;2)(3;4)5);6). What
should a cumulative ACK of object 4 confirm? The most logical
definition would say it confirms receipt of object 4 and all objects
that precede 4 in the partial order, in this case, object 3. Nothing
is said about the arrival of objects 1 or 2. With this alternative
interpretation where cumulative ACKs depend on the partial order, the
sender must examine the partial order matrix to determine which
buffers can be released. In this example, scanning column 4 of the
matrix reveals that object 3 must come before object 4 and therefore
both object buffers (and any buffers from a previous period) can be
released.
Other partial order acknowledgment policies are possible for a
protocol providing a partial order service including the use of
selective ACKs (which has been proposed in [JB88] and implemented in
the Cray TCP [Chang93]) as well as the current TCP strategy where an
ACK of i also ACKs everything <= i (in a cyclical sequence number
space). The authors are investigating an ACK policy which utilizes a
combination of selective and "partial-order-cumulative"
acknowledgments. This is accomplished by replacing the current TCP
cumulative ACK with one which has the partial order meaning as
described above and augmenting this with intermittent selective ACKs
when needed.
In another area, the notion of fragmented delivery, mentioned in the
beginning of Section 4, looks like a promising technique for certain
classes of applications which may offer a substantial improvement in
memory utilization. Briefly, the term fragmented delivery refers to
the ability to transfer less-than-complete objects between the
transport layer and the user application (or session layer as the
case may be). For example, a 1Mbyte object could potentially be
delivered in multiple "chunks" as segments arrive thus freeing up
valuable memory and reducing the delay on those pieces of data. The
scenario becomes somewhat more complex when multiple "parallel
streams" are considered where the application could now receive
pieces of multiple objects associated with different streams.
Additional work in the area of implementing a working partial order
protocol is being performed both at the University of Delaware and at
the LAAS du CNRS laboratory in Toulouse, France - particularly in
support of distributed, high-speed, multimedia communication. It will
be interesting to examine the processing requirements for an
implementation of a partial order protocol at key events (such as
object arrival) compared with a non-partial order implementation.
Finally, the authors are interested in the realization of a network
application utilizing a partial order service. The aim of such work
is threefold: (1) provide further insight into the expected
performance gains, (2) identify new issues unique to partial order
transport and, (3) build a road-map for application designers
interested in using a partial order service.
7. Summary
This RFCintroduces the concepts of a partial order service and
discusses the practical issues involved with including partial
ordering in a transport protocol. The need for such a service is
motivated by several applications including the vast fields of
distributed databases, and multimedia. The service has been
presented as a backward-compatible extension to TCP to adapt to
applications with different needs specified in terms of QOS
parameters.
The notion of a partial ordering extends QOS flexibility to include
object delivery, reliability, and temporal value thus allowing the
transport layer to effectively handle a wider range of applications
(i.e., any which might benefit from such mechanisms). The service
profile described in Section 4 accurately characterizes the QOS for a
partial order service (which encompasses the two extremes of total
ordered and unordered transport as well).
Several significant modifications have been proposed and are
summarized here:
(1) Replacing the requirement for ordered delivery with one for
application-dependent partial ordering
(2) Allowing unreliable and partially reliable data transport
(3) Conducting a non-symmetrical connection (not entirely foreign
to TCP, the use of different MSS values for the two sides
of a connection is an example)
(4) Management of "objects" rather than octets
(5) Modified acknowledgment strategy
(6) New definition for the send and receive "windows"
(7) Extension of the User/TCP interface to include certain
QOS parameters
(8) Use of new TCP options
As evidenced by this list, a partial order and partial reliability
service proposes to re-examine several fundamental transport
mechanisms and, in so doing, offers the opportunity for substantial
improvement in the support of existing and new application areas.
8. References
[ACCD93a] Amer, P., Chassot, C., Connolly, T., and M. Diaz,
"Partial Order Transport Service for Multimedia
Applications: Reliable Service", Second International
Symposium on High Performance Distributed Computing
(HPDC-2), Spokane, Washington, July 1993.
[ACCD93b] Amer, P., Chassot, C., Connolly, T., and M. Diaz,
"Partial Order Transport Service for Multimedia
Applications: Unreliable Service", Proc. INET '93, San
Francisco, August 1993.
[AH91] Anderson, D., and G. Homsy, "A Continuous Media I/O
Server and its Synchronization Mechanism", IEEE
Computer, 24(10), 51-57, October 1991.
[AS93] Agrawala, A., and D. Sanghi, "Experimental Assessment
of End-to-End Behavior on Internet," Proc. IEEE INFOCOM
'93, San Francisco, CA, March 1993.
[BCP93] Claffy, K., Polyzos, G., and H.-W. Braun, "Traffic
Characteristics of the T1 NSFNET", Proc. IEEE INFOCOM
'93, San Francisco, CA, March 1993.
[Bol93] Bolot, J., "End-to-End Packet Delay and Loss Behavior
in the Internet", SIGCOMM '93, Ithaca, NY, September
1993.
[CAC93] Conrad, P., Amer, P., and T. Connolly, "Improving
Performance in Transport-Layer Communications Protocols
by using Partial Orders and Partial Reliability",
Work in Progress, December 1993.
[Chang93] Chang, Y., "High-Speed Transport Protocol Evaluation --
the Final Report", MCNC Center for Communications
Technical Document, February 1993.
[Dee89] Deering, S., "Host Extensions for IP Multicasting," STD
5, RFC1112 Stanford University, August 1989.
[DS93] Diaz, M., and P. Senac, "Time Stream Petri Nets: A
Model for Multimedia Synchronization", Proceedings of
Multimedia Modeling '93, Singapore, 1993.
[HKN91] Hardt-Kornacki, S., and L. Ness, "Optimization Model
for the Delivery of Interactive Multimedia Documents",
In Proc. Globecom '91, 669-673, Phoenix, Arizona,
December 1991.
[JB88] Jacobson, V., and R. Braden, "TCP Extensions for
Long-Delay Paths", RFC1072, LBL, USC/Information
Sciences Institute, October 1988.
[JBB92] Jacobson, V., Braden, R., and D. Borman, "TCP
Extensions for High Performance", RFC1323, LBL, Cray
Research, USC/Information Sciences Institute, May 1992.
[LMKQ89] Leffler, S., McKusick, M., Karels, M., and J.
Quarterman, "4.3 BSD UNIX Operating System",
Addison-Wesley Publishing Company, Reading, MA, 1989.
[OP91] O'Malley, S., and L. Peterson, "TCP Extensions
Considered Harmful", RFC1263, University of Arizona,
October 1991.
[Pos81] Postel, J., "Transmission Control Protocol - DARPA
Internet Program Protocol Specification," STD 7,
RFC793, DARPA, September 1981.
Security Considerations
Security issues are not discussed in this memo.
Authors' Addresses
Tom Connolly
101C Smith Hall
Department of Computer & Information Sciences
University of Delaware
Newark, DE 19716 - 2586
EMail: connolly@udel.edu
Paul D. Amer
101C Smith Hall
Department of Computer & Information Sciences
University of Delaware
Newark, DE 19716 - 2586
EMail: amer@udel.edu
Phill Conrad
101C Smith Hall
Department of Computer & Information Sciences
University of Delaware
Newark, DE 19716 - 2586
