Re: Problems with Solaris 2.6 and Squid-1.1.20 !

> On Wed, 11 Mar 1998, Tomi Orava wrote:
>
> > Currently when running 2.5.1, the lod is around 0.3 when there are
> > around 200
> > users using the proxy. However, if I try run the same software with the
> > same
> > hardware with Solaris 2.6, the CPU usage goes through the roof (more
> > than 1.xx).
> >
> > Do people have similar setups working OK with Solaris 2.6 and Squid
> > 1.1.20 ?
>
> Yes, we have (had, now it's a 1.2) a 1.1.20 running on 2.6 with >1
> mill. requests/day, so it's not a general 'incompatibility'.
>
> Overall system load, however, is a bad indicator for a system's
> performance.
> It just counts the average number of ready-to-run-processes.
> Try checking with vmstat, iostat or better yet use the SymbEL package
> (available on www.sun.com) to see what's really causing the load.

Well, to me it seems that a test machine (Sun SS5) running Solaris 2.6 and
Squid-1.1.20+retry is under heavier load than the similar machine in
production use (also SS5/200MB/8GB/Sol 2.5.1). Here are short outputs
from the cache_manager:

Solaris 2.5.1 (Production machine):
-----------------------------------

Squid Object Cache: Version 1.1.20
Start Time: Thu, 12 Mar 1998 02:39:26 GMT
Current Time: Thu, 12 Mar 1998 12:58:43 GMT
Connection information for squid:
        Number of TCP connections: 28425
        Number of UDP connections: 0
        Connections per hour: 2754.0
        Select loop called: 485276 times, 76.569 ms avg
Cache information for squid:
        Storage Swap size: 3838 MB
        Storage Mem size: 6026 KB
        Storage LRU Expiration Age: 60.00 days
        Requests given to unlinkd: 0
        Unused fileno stack count: 0
Resource usage for squid:
        CPU Time: 1513 seconds (952 user 561 sys)
        CPU Usage: 4%
        Maximum Resident Size: 0 KB
        Page faults with physical i/o: 188190

Solaris 2.6 (Test machine):
------------------------------

Squid Object Cache: Version 1.1.20+retry
Start Time: Thu, 12 Mar 1998 12:48:14 GMT
Current Time: Thu, 12 Mar 1998 13:03:17 GMT
Connection information for squid:
        Number of TCP connections: 65
        Number of UDP connections: 354
        Connections per hour: 1670.4
        Select loop called: 1275 times, 708.235 ms avg
Cache information for squid:
        Storage Swap size: 0 MB
        Storage Mem size: 0 KB
        Storage LRU Expiration Age: 30.00 days
        Requests given to unlinkd: 0
        Unused fileno stack count: 0
Resource usage for squid:
        CPU Time: 565 seconds (563 user 2 sys)
        CPU Usage: 63%
        Maximum Resident Size: 0 KB
        Page faults with physical i/o: 7

----------------------------------------------------

Notice the big difference between the selection loop average times as
well as in CPU usage values. Vmstat command verifies that it really
is the case that Squid is using 60-90 % of CPU from time to time, which is
abnormal behaviour.

BTW. The Solaris 2.6 test machine has all the latest patches which are included
in the March/98 patch cluster available from Sunsolve.

//Tomi Orava

Tomi Orava | The views and opinions expressed in this article
Oy LM Ericsson Ab | are those of the poster and not his employer.
LMF/ZT/Unix Administration
phone: +358 9 2992604
Received on Thu Mar 12 1998 - 05:36:42 MST