一、 概述
squid是一款优秀的open source的代理服务器软件，可以运行于多种系统平台上，但是同其他商业化的产品相比，其缺点也是很明显的，那就是它的命中率和效率相对低下。
本文主要讨论在不改变硬件条件之下，通过以下手段对squid进行性能优化：
1． 编译一个高效的、精简的新内核；
2． 对Cache分区采用reiserfs日志文件系统；
3． 重新编译squid；
4． 优化squid配置；
对于优化的效果，我使用大名鼎鼎的cacheflow公司的测试工具——cfmc进行测试比较：
首先用cfmc自带的一个脚本，从squid.conf的access.log中抽出RUL，然后以这些RUL为依据进行测试。
服务器配置：
型号：HP LH3
CPU：PII450
内存：256MB 100MHz ECC SDRAM DIMM
硬盘：9.1GB Hot-Swap Ultra2 Hard Disk
网卡：Ethernet Express PRO 100 10/100M X 2
操作系统：RedHat 7.1
优化前测试结果：
Iteration 0: Cumulative statistics; 933 seconds elapsed
Total objects: 72599, total object size: 513211102 bytes
Average object size: 7069 bytes
Average object response time: 2707 milliseconds
Objects per second: 77.81
Bytes per second: 550065, min: 550065, max: 879873
URLs discarded due to socket or connection failures: 6955
Redirections: 1017, Cookied objects: 1036
Pragma no-cache objects: 1656, Non-200 HTTP response codes: 2505
优化后测试结果：
Iteration 0: Cumulative statistics; 688 seconds elapsed
Total objects: 72599, total object size: 403833100 bytes
Average object size: 5562 bytes
Average object response time: 1890 milliseconds
Objects per second: 105.52
Bytes per second: 586966, min: 586966, max: 995582
URLs discarded due to socket or connection failures: 16372
Redirections: 1658, Cookied objects: 1000
Pragma no-cache objects: 1454, Non-200 HTTP response codes: 3132
通过对以上测试结果的比较，我们可以看出：
优化后的Objects per second增加了35.6％，Average object response time减少了43.2％，对整体性能的提升还是比较明显的。
二、 编译新内核
我们采取这样的方式来编译内核：取消内核的模块支持，将服务器所有的硬件驱动编译到内核中，此外还要注意将对reiserfs文件系统的支持也编译到内核中，从而在提供系统性能的同时增加系统安全性。在2.4.10以上的版本中，已经内置了对reiserfs文件系统的支持。
首先从http://www.kernel.org/pub/linux/kernel/v2.4/下载linux-2.4.12.tar.gz，这是当前最新的内核。
然后解开内核文件：
tar xvzf linux-2.4.12.tar.gz
进入新生成的目录并执行：
cd linux
make mrproper
make config
根据我的服务器的配置，我选择这样编译内核：
* Code maturity level options
Prompt for development and/or incomplete code/drivers (CONFIG_EXPERIMENTAL) [Y/n/?]
* Loadable module support
Enable loadable module support (CONFIG_MODULES) [N/y/?]
* Processor type and features
Processor family (386, 486, 586/K5/5x86/6x86/6x86MX, Pentium-Classic, Pentium-MM
X, Pentium-Pro/Celeron/Pentium-II, Pentium-III/Celeron(Coppermine), Pentium-4, K
6/K6-II/K6-III, Athlon/Duron/K7, Crusoe, Winchip-C6, Winchip-2, Winchip-2A/Winch
ip-3, CyrixIII/C3) [Pentium-Pro/Celeron/Pentium-II]
* General setup
Networking support (CONFIG_NET) [Y/n/?]
PCI support (CONFIG_PCI) [Y/n/?]
PCI access mode (BIOS, Direct, Any) [Any]
defined CONFIG_PCI_GOANY
PCI device name database (CONFIG_PCI_NAMES) [Y/n/?]
System V IPC (CONFIG_SYSVIPC) [Y/n/?]
Sysctl support (CONFIG_SYSCTL) [Y/n/?]
Kernel core (/proc/kcore) format (ELF, A.OUT) [ELF]
defined CONFIG_KCORE_ELF
* Plug and Play configuration
Plug and Play support (CONFIG_PNP) [Y/n/?]
* Block devices
Normal PC floppy disk support (CONFIG_BLK_DEV_FD) [Y/n/?]
* Networking options
Kernel/User netlink socket (CONFIG_NETLINK) [Y/n/?]
Routing messages (CONFIG_RTNETLINK) [Y/n/?]
Network packet filtering (replaces ipchains) (CONFIG_NETFILTER) [Y/n/?]
Unix domain sockets (CONFIG_UNIX) [Y/n/?]
TCP/IP networking (CONFIG_INET) [Y/n/?]
IP: advanced router (CONFIG_IP_ADVANCED_ROUTER) [Y/n/?]
IP: policy routing (CONFIG_IP_MULTIPLE_TABLES) [Y/n/?]
IP: use netfilter MARK value as routing key (CONFIG_IP_ROUTE_FWMARK) [Y/n/?]
IP: fast network address translation (CONFIG_IP_ROUTE_NAT) [Y/n/?]
IP: equal cost multipath (CONFIG_IP_ROUTE_MULTIPATH) [Y/n/?]
* IP: Netfilter Configuration
Connection tracking (required for masq/NAT) (CONFIG_IP_NF_CONNTRACK) [Y/n/?]
FTP protocol support (CONFIG_IP_NF_FTP) [Y/n/?]
IP tables support (required for filtering/masq/NAT) (CONFIG_IP_NF_IPTABLES) [Y/n/?]
limit match support (CONFIG_IP_NF_MATCH_LIMIT) [Y/n/?]
Multiple port match support (CONFIG_IP_NF_MATCH_MULTIPORT) [Y/n/?]
Connection state match support (CONFIG_IP_NF_MATCH_STATE) [Y/n/?]
Packet filtering (CONFIG_IP_NF_FILTER) [Y/n/?]
Full NAT (CONFIG_IP_NF_NAT) [Y/n/?]
REDIRECT target support (CONFIG_IP_NF_TARGET_REDIRECT) [Y/n/?]
LOG target support (CONFIG_IP_NF_TARGET_LOG) [Y/n/?]
ATA/IDE/MFM/RLL support (CONFIG_IDE) [Y/n/?]
* IDE, ATA and ATAPI Block devices
Enhanced IDE/MFM/RLL disk/cdrom/tape/floppy support (CONFIG_BLK_DEV_IDE) [Y/n/?]
Include IDE/ATA-2 DISK support (CONFIG_BLK_DEV_IDEDISK) [Y/n/?]
Use multi-mode by default (CONFIG_IDEDISK_MULTI_MODE) [Y/n/?]
Include IDE/ATAPI CDROM support (CONFIG_BLK_DEV_IDECD) [Y/n/?]
* SCSI support
SCSI support (CONFIG_SCSI) [Y/n/?]
* SCSI support type (disk, tape, CD-ROM)
SCSI disk support (CONFIG_BLK_DEV_SD) [Y/n/?]
Maximum number of SCSI disks that can be loaded as modules (CONFIG_SD_EXTRA_DEVS) [8]
* Some SCSI devices (e.g. CD jukebox) support multiple LUNs
Enable extra checks in new queueing code (CONFIG_SCSI_DEBUG_QUEUES) [Y/n/?]
* SCSI low-level drivers
AMI MegaRAID support (CONFIG_SCSI_MEGARAID) [Y/n/?]
SYM53C8XX SCSI support (CONFIG_SCSI_SYM53C8XX) [Y/n/?]
default tagged command queue depth (CONFIG_SCSI_NCR53C8XX_DEFAULT_TAGS) [4]
maximum number of queued commands (CONFIG_SCSI_NCR53C8XX_MAX_TAGS) [32]
synchronous transfers frequency in MHz (CONFIG_SCSI_NCR53C8XX_SYNC) [80]
* Network device support
Network device support (CONFIG_NETDEVICES) [Y/n/?]
* Ethernet (10 or 100Mbit)
Ethernet (10 or 100Mbit) (CONFIG_NET_ETHERNET) [Y/n/?]
EISA, VLB, PCI and on board controllers (CONFIG_NET_PCI) [Y/n/?]
EtherExpressPro/100 support (CONFIG_EEPRO100) [Y/n/?]
* Input core support
Input core support (CONFIG_INPUT) [Y/n/?]
Keyboard support (CONFIG_INPUT_KEYBDEV) [Y/n/?]
* Character devices
Virtual terminal (CONFIG_VT) [Y/n/?]
Support for console on virtual terminal (CONFIG_VT_CONSOLE) [Y/n/?]
Standard/generic (8250/16550 and compatible UARTs) serial support (CONFIG_SERIAL) [Y/n/?]
Unix98 PTY support (CONFIG_UNIX98_PTYS) [Y/n/?]
Maximum number of Unix98 PTYs in use (0-2048) (CONFIG_UNIX98_PTY_COUNT) [8]
* File systems
Reiserfs support (CONFIG_REISERFS_FS) [Y/n/?]
ISO 9660 CDROM file system support (CONFIG_ISO9660_FS) [Y/n/?]
/proc file system support (CONFIG_PROC_FS) [Y/n/?]
/dev/pts file system for Unix98 PTYs (CONFIG_DEVPTS_FS) [Y/n/?]
* Console drivers
VGA text console (CONFIG_VGA_CONSOLE) [Y/n/?]
这里需要说明的是：如果要使用透明代理模式，要仔细选择Netfilter，因为我们还得用iptables来重定向tcp包，以满足透明代理的要求。
假如您的服务器配置与我的不同，只需要更改相应的硬件配置参数即可，比如CPU类型、网卡、SCSI硬盘等。
接着：
make bzImage
完成后，将 arch/i386/boot/目录下的新内核bzImage拷贝到/boot/目录下，并更名为opt：
cp arch/i386/boot/bzImage /boot/
mv /boot/bzImage /boot/opt
然后编辑lilo.conf如下所示：
boot=/dev/sda
map=/boot/map
install=/boot/boot.b
prompt
timeout=50
message=/boot/message
default=opt
image=/boot/vmlinuz-2.4.2-2
label=linux
initrd=/boot/initrd-2.4.2-2.img
read-only
root=/dev/sda1
image=/boot/opt
label=opt
read-only
root=/dev/sda1
最后执行：
# lilo
Added linux
Added opt *
好了，这样我们就可以用新内核启动了，您可以发现，这样自己编译的内核要小的多，也快得多。
