Re: [squid-users] 2.6-stable3 performance weirdness with polymix-4 testing

From: Pranav Desai <pranavadesai@dont-contact.us>
Date: Mon, 28 Aug 2006 18:57:33 -0700

On 8/28/06, Adrian Chadd <adrian@creative.net.au> wrote:
> On Mon, Aug 28, 2006, Pranav Desai wrote:
>
> > >A single drive and you expect 1000 req/s? Is this a solid-state drive or
> > >something with infinitely small seek time?
> > >
> > No, its just a regular drive. I am already working on that, but its
> > kind of difficult to get things changed quickly :-).
> > But the good part is that even with 1 disk its able to do 1000 req/s
> > atleast for the first 6 hrs, so I am sure I can push it more once I
> > get some better hardware.
>
> I think you'll find that the performance is degrading because:
>
> * (a) the disk write queues are slowly filling up and taking longer to write;
> * (b) the disk is filling up and fragmentation+object replacement+non-linear
> file allocation kicks in.
>
> Faster hardware won't fix that. More disks may. COSS will help a lot for
> the smaller objects. But a lot more work needs to be done to improve the
> Squid disk store under load.
>

for (a), shouldnt the disk queue also affect the performance during
the first phase of the test. The first phase is almost perfect with no
degradation in times.
If I may send you the graphs, in which it seems very wrong that the
degradation happens so suddenly.
Maybe I will try a few test with a longer first phase to see if the
disk queue catches up and degrades the performance.

I will be trying COSS shortly. Even I am expecting it to do much
better. What other kind of work do you have in mind. Another
filesystem or something ?
I am thinking of giving this http://logfs.sourceforge.net/ a try.
But I dont know if it needs changes in squid to make efficient use of
the filesystem.

> > It didnt in the first phase. In fact it did very well in the first
> > phase compared to the first phase of 2.5, but somehow after the idle
> > phase it just wasnt able to recover and the polybench wasnt even able
> > to push to 1000 req/s because the response times were so high.
>
> Whats your squid.conf look like? It sounds like the disks just couldn't
> keep up with your request rate.
>

visible_hostname 10.51.6.102
cache_dir aufs /var/cache/squid 30720 16 256
http_port 8080
request_body_max_size 0
snmp_port 3401
negative_ttl 0 minutes
pid_filename /var/run/squid.pid
coredump_dir /var/log/squid
cache_effective_user squid
cache_effective_group squid
cache_access_log /var/log/squid/access.log
cache_log /var/log/squid/cache.log
cache_store_log none
cache_swap_log /var/log/squid/swap.log
logfile_rotate 10
icp_port 3130
icp_query_timeout 20000
log_icp_queries on
extension_methods SEARCH PROPPATCH
forwarded_for on
acl all src 0.0.0.0/0.0.0.0
acl localhost src 127.0.0.1 10.51.6.102
acl manager proto cache_object
acl snmppublic snmp_community public
http_access allow localhost
miss_access allow all
http_access allow all
snmp_access allow snmppublic all
memory_pools off

>
>
>
> Adrian
>
>

-- 
------------------------------
http://pd.dnsalias.org
Received on Mon Aug 28 2006 - 19:57:35 MDT

This archive was generated by hypermail pre-2.1.9 : Fri Sep 01 2006 - 12:00:02 MDT