Re: [squid-users] running out of file descriptors (SOLVED!)

Thanks Henrik. Unless there is something more you think I need to be
alert for because of the changes - this problem is solved. I've
responded to various questions you last asked inline below.

Thanks!

Henrik Nordstrom wrote:
>
> ons 2003-04-02 klockan 23.05 skrev Mike Rambo:
>
> > > a) Is it swapping?
> >
> > Yes, a little anyway. Since we rebooted yesterday is is about 5MB into
> > swap. Prior to the reboot it had reached almost 100MB swap after
> > approximately 30 days or so uptime.
>
> The amount into swap is not as important as the current swapping
> activity.
>
> If you use Linux see the output of
>
> vmstat 5
>

I checked this dozens of times throughout the day. Zero's every time I
checked .

> If you see more than occasional activity in the "si" and "so" columns
> then your system is swapping.
>
> Other OS:es can be inspected similarly, but the columns may differ
> slightly. See the documentation for vmstat, iostat or sar, depending on
> platform and taste.
>
> > cache_dir diskd /mnt/cache 11000 16 256
>
> One single drive?
>

There are three UW SCSI3 drives in the machine. One 18GB drive is
dedicated to the cache. It is formatted reiser fs.

>
> > client_http.requests = 69.485624/sec
> > client_http.hits = 30.510768/sec
>
> Fairly medium traffic then.
>
> > File descriptor usage for squid:
> > Maximum number of file descriptors: 1024
> > Largest file desc currently in use: 930
> > Number of file desc currently in use: 637
> > Files queued for open: 0
> > Available number of file descriptors: 387
> > Reserved number of file descriptors: 100
> > Store Disk files open: 0
> >
> > Seems like I am running out of descriptors just as the log says.
>
> Maybe. Increasing the number of filedescriptors will probably help.
>
> > What other things can I look at to see why?
>
> You can look into cachemgr filedescriptor utilization to figure out what
> all the filedescriprots are being used for.
>
> It is also interesting to look into the service times. If your service
> times for processing cache hits is high then there most likely is other
> problems.
>
> You can also try the following settings to reduce the amount of
> filedescriptors used:
>
> half_closed_clients off
>
> And if that does not help
>
> server_persistent_connections off
>
> and finally if still a problem
>
> client_persistent_connections off

I tried all three of these. The first had no discernible effect. I tried
both of the _persisent_connections options this morning together. This
has been a HUGE win. The file descriptors problem went away and
interactivity (read speed of access to web pages) has went up very, very
noticeably. I can't hardly believe how much faster it's running. I
noticed the requests per second we're servicing has went up by at least
25% and really it's up almost by a third. We're no longer showing
anywhere near full useage of file descriptors - the vast majority are
free.

Is there a downside to these changes? Are there any other things I need
to be on the alert for now that I've made them?

>
> Regards
> Henrik
>

Once again. Thanks very much for the help.

-- 
Mike Rambo
mrambo@lsd.k12.mi.us