RE: [squid-users] TCP_MEM_HIT performance

Thanks Henrik for your response. See my comments below:

 
> On Wed, 7 Apr 2004, Francisco Lopez wrote:
>
> > Tests are performed using Jmeter 2.0 from 3 different pc's
> (to avoid
> > client stress). Each test run is a constant number of
> requests (3000)
> > varying the number of simultaneous users (3, 6, 15, 30, 60,
> 150). The
> > objects requests are chosen on each loop among a set of 10, with an
> > average page size of 50k.
>
> 3000 * 50k = ~150 MB. This work set easily fits in memory.

There are 3000 requests, however only 10 different pages. So the total
size of objects in memory should be 500k. Anyway, this is not important.
>
...
> > c) Filesystem type affects performance (ext3 w/journaling is slower
> > than non journaling filesystems)
>
> With no significant disk I/O there is not much impact of filesystem..

I don't agree with this point. Since:
A) there are (on the top users test) 150 simultaneous users
B) squid doesn't put in memory objects not received from the network

All of those requests 150 (ok, not all of the users are simultaneously
requesting pages but the number should be around that) must be read from
disk. This is pretty high i/o and filesystem performance should be a
factor.

...

> > BUT
> > a) AUFS performs the same or slower than UFS. According to
> what I've
> > been reading, AUFS should take advantage of multithreading for I/O,
> > and in high loads this should improve i/o performance
>
> See above.

Again, I think that on high load AUFS should perform better and 150
users sending requests is generating a lot of i/o on the server (unless
the store manager is keeping some kind of memory pool, in that case I
agree that even when there are 150 requests only 10 of them are unique;
since there are only 10 different pages in my test).
>
> > b) 100% TCP_MEM_HIT performs the same or slower than TCP_HIT
>
> Then you are swapping. Make sure your system is not swapping.
> Assuming you
> have sufficient ram you may need to disable the swap partition.
>

Server is not swapping. It has 512MB ram and has memory available.

> > So my questions are:
> > Why AUFS is slower than UFS ?
>
> Because it sacrifies CPU time to utilise disk I/O seek time,
> which is not
> a bottleneck for you as you don't use the disks with such small
> workingset.

According to the tests published in the Squid book by Duane Wessels,
AUFS should be faster. My system has 2 PIII, 1Ghz, and I don't think CPU
is an issue here.
Or is slower because of the cache magnitude you say ?

>
> > Why memory hits are slower than disk hits ?
>
> Because something is terribly wrong in your setup.
>

Any ideas of what could be wrong ?

I know that without the configuration file is very hard to guess, but
perhaps you've seen this before.

Francisco
Received on Wed Apr 07 2004 - 18:09:08 MDT