savannah-hackers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Savannah-hackers] Re: Savannah CVS update performance...


From: Eric Blossom
Subject: [Savannah-hackers] Re: Savannah CVS update performance...
Date: Thu, 20 Nov 2003 10:03:23 -0800
User-agent: Mutt/1.4i

On Thu, Nov 20, 2003 at 04:27:56PM +0100, Mathieu Roy wrote:
> Vincent Caron <address@hidden> a tapot? :
> 
> > Eric Blossom wrote:
> >> OK.  I just reproduced the same test on the same machine, I'm now
> >> down
> >> to 6.3 seconds.  Can you say highly variable...
> >
> > Apologies, Savannah does have serious slow down problems. It's peak
> > time now, and a gnuradio checkout took me 5m40 locally on the server !
> >
> >  From here, I can see kswapd taking most time, half of the processes
> > in the D state (uninterruptible state = low-level I/O bottleneck I
> > guess), and no a byte of free swap. Conclusion : kswapd is swapping in
> > and out permanently, this machine is in dire of swap space. It only
> > has a 340MB swap for 1GB RAM. The Linux 2.4 VM requires 2GB of swap
> > for proper operation on this hardware.
> >
> > I higly recommend to enlarge the current swap partition or add a 1,5GB
> > one. Someone is handling this ?
> >
> > PS: adding some RAM wouldn't hurt, if you have some spare, please feed
> > it to subversions. It would at least help to reduce the time periods
> > where swap is awfully degrading performances.
> >
> 
> The swap is almost unused.
> 
> address@hidden:~> vmstat
>    procs                      memory    swap          io     system         
> cpu
>  r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  
> id
>  1 29  1 353420   7400  46504 318184   2   0     0     0    0     0   0   1   
> 1
> 
> 
> So this can't be the problem.

I'm not a vmstat wizard, but the swpd entry seems to indicate that all
of the swap, the 340M, is committed.  You've got 29 processes blocked,
though there's not much i/o being reported.

If I was near the machine, I'd take a look at the disk light and see
if it's on all the time.  Is there a command that gives average i/o to
a disk that includes paging plus normal file system activity?

What kind of disks are on the system?
Any kind of raid?

FWIW, I can sustain over 50 MB/sec on one of my servers using two
inexpensive Western Digital 120GB drives and the LVM in a RAID 0
configuration. 

With three SCSI drives I can sustain 115 MB/sec, but of course they
cost a lot.

> But one more time I see a anoncvs process taking 684 MB of RAM. This
> is not the normal behavior of CVS.
> 
> This must be the problem.

Seems awfully large, and worth investigating.

I'd also add or enlarge the swap partition.

Eric




reply via email to

[Prev in Thread] Current Thread [Next in Thread]