Performance Problems.. Server hardware smoked by $500 box?

John Straiton jsmailing at clickcom.com
Thu Sep 11 14:36:20 PDT 2003


> > #bonnie -s 1    (run on NFS share)
> >            -------Sequential Output-------- ---Sequential 
> Input---  --Random--
> >            -Per Char- --Block--- -Rewrite-- -Per Char-  --Block---  
> > --Seeks--- Machine MB K/sec %CPU K/sec %CPU K/sec %CPU 
> K/sec %CPU  K/sec %CPU  K/sec %CPU
> > PROD     1  6364 25.0  9121  4.9   166  0.3  2906 13.5  
> 94108 28.6  398.8  3.3
> > DEV      1 11246 65.3 11259  6.4  6733  4.3 11009 41.9 
> 292571 95.3 5612.2 28.9
> 
> Did you build these tables in a variable font or something?  
> I had to do a lot of massaging to get them to line up and be 
> readable in mutt and vi.
Damn outlook.... =(  Cut n pasted from SecureCRT.


> Using less CPU% on the PROD box without memory pressure says 
> I/O issues to me.
> 
> Is your SCSI bus clean?  Do you get any SCSI errors?   
> Is write cache enabled on the SCSI?
> How big is the hardware cache on the SCSI? and the IDE?
> ...etc..
Both bonnie scores are to the same place... To the DB/File Server's
RAID5 chassis via NFS. Since nothing is being served (even php session
files are written over NFS), the local drive performance shouldn't be an
issue in my mind, tho' I included it for completeness. Please correct me
if I'm wrong.

> Have you tried putting the DEV NIC in the PROD machine?
Can't. The production machine is a Dell 1650, a 1 Unit machine with zero
PCI slots (at least ones that a 3c905 will fit in). We were gonna try
that yesterday *frown*.
 
> Have you watched systat -vm 1 while testing?  Maybe you can 
> tell if you are getting an interupt storm or something.
Will try..
 
> Do you have a lot of apache processes in biord during the run 
> on the PROD box?
Yes, in all cases the # of Min/Max/Start servers on the production
machine has been approximately double over the development. But
everything I've read seemed to indicate that this *helps* serving speed
until you run into vm issues. Am I misinformed?
 
> Are the filesystems mounted with the same options on 
> production and development systems?
> 
> > Recv   Send    Send                          
> > Socket Socket  Message  Elapsed              
> > Size   Size    Size     Time     Throughput  
> > bytes  bytes   bytes    secs.    10^6bits/sec  
> >  57344  32768  32768    10.01      92.74 # DB FROM PROD  
> >  57344  32768  32768    10.00      87.61 # DB FROM DEV
> 
> Either machine should be able to saturate a 100Mbps link if 
> the data doesn't have to come from disk first.
My thoughts too. This test was run using netperf which to the best of my
knowledge, just runs 1's and 0's down the link (client/server) to test
speed, it doesn't actually try sending a file or anything.

John Straiton
jks@ clickcom.com
Clickcom, Inc
704-365-9970x101 




More information about the freebsd-performance mailing list