> All the caches I've been told about thus far have a HTTP IO read
> peak at around 1k to 2k.
>
> I'm going to try and add client-side read/write histograms to
> see whats going on there, but before I do - could people please
> take a look at their cachemgr output and see whether their read
> histogram looks somewhat like this? I'm very interested if
> the read histogram looks like this and you're running a HTTP
> accelerator (ie, with the -servers- close to Squid, with a low
> RTT like ethernet.)
If loopback is also fine, then here it is, straight from the wiki rproxy:
HTTP I/O
number of reads: 2421779
Read Histogram:
1- 1: 44119 2%
2- 2: 6377 0%
3- 4: 246467 10%
5- 8: 72953 3%
9- 16: 141185 6%
17- 32: 286742 12%
33- 64: 668808 28%
65- 128: 258054 11%
129- 256: 262559 11%
257- 512: 70720 3%
513- 1024: 52099 2%
1025- 2048: 67733 3%
2049- 4096: 91963 4%
4097- 8192: 93737 4%
8193-16384: 43378 2%
16385-32768: 9924 0%
--
/kinkie
