Re: the performance of squid
早上起来，看到一个讨论squid性能的邮件，有很多问题都是一直困挠我的。不过，读完邮件后，释怀不少了。
现将部分邮件帖出如下的：
Hi, I use squid as a web accelerating cache proxy for the backend MS-IIS-6.
When the requests per-second reache 2000, the squid will eat CPU
resources to 99%. The files are some small size image, css, etc. I fill
the regex computing or other operation may result in the dropping of
the squid's performance. Is there some other way to improve squid's efficiency ?
My computer hardware: 3.0 G Interl CPU, 4G memory, 100G SCSI disk.
从邮件正文，可以发现这位用户使用的squid在处理小对象的时候，出现了性能瓶颈。对于频繁的访问小对象，在对header的处理势必要耗费大量的cpu时间。从我的角度来看，要想提高其性能，应该尽量减少对header的处理。
读到Adrian Chadd的回复，他也是建议读者使用改写后的squid-2-Head。可见在head这一模块，性能还是有很大的提升空间。下面是Adrian Chadd回复的部分邮件内容：
I'm working on identifying and improving the performance of Squid-2 HEAD.
You could give that a whirl and let me know how it goes. I'd appreciate
any testing that you could give it.
最后问题提出者通过修改了部分配置文件，提高了squid的性能。其做法如下：
1) If requets/second is high, and the files be to cached are small size and many
number, the configure items: cache_mem, memory_pools should be set to use,
and client_persistent_connections should be set 'on' when you use squid-2.6.X
(in which the system's epoll be used);
2)If request/second are not too large(such as 1000), and the files
data be cached are too large(such as 100G ), we should close the memory_pools and set
cache_mem to be a lower value(such as 256MB), this will let the OS system
to use more ram memory and use swap memory fewer, and this will decrease
the page-faults of squid very much and squid will not be blocked on DISK I/O.
可以看出，第一种方案之所以是成功的，是通过合理的使用了cache_mem, memory_pools减少了对Header的处理次数。而使用持久连接，避免了过于频繁的tcp连接和关闭。Tcp的连接和关闭，都是需要耗费系统资源的。
对于第二种情况的话，使用cache_mem、 memory_pools的话，是要耗费过多的内存的，明显是不可取的。读者这样做，避免了这个问题。至于是否真的DISK I/O不阻塞了，那就得通过测试才能确定了。
以上两种解决方法，都是非常具有参考价值的。
