[CentOS] Adding RAM

Thu Dec 18 17:24:32 UTC 2008
John <jses27 at gmail.com>

> -----Original Message-----
> From: centos-bounces at centos.org 
> [mailto:centos-bounces at centos.org] On Behalf Of Matt
> Sent: Thursday, December 18, 2008 10:37 AM
> To: CentOS mailing list
> Subject: Re: [CentOS] Adding RAM
> 
> >> A bit of bottle neck.
> >>
> >> Device:    rrqm/s wrqm/s   r/s   w/s  rsec/s  wsec/s    
> rkB/s    wkB/s
> >> avgrq-sz avgqu-sz   await  svctm  %util
> >> sda          0.38 176.63 70.32 78.26  813.46 2044.82   
> 406.73  1022.41
> >>   19.24     0.40   19.17   4.04  60.07
> >> sda1         0.00   0.00  0.00  0.00    0.00    0.00     
> 0.00     0.00
> >>    5.28     0.00   23.61  19.33   0.00
> >> sda2         0.38 176.63 70.32 78.26  813.45 2044.82   
> 406.73  1022.41
> >>   19.24     0.40   19.17   4.04  60.07
> >> dm-0         0.00   0.00 70.71 255.60  813.45 2044.82   406.73
> >> 1022.41     8.76     2.90    8.87   1.84  60.10
> >> dm-1         0.00   0.00  0.00  0.00    0.00    0.00     
> 0.00     0.00
> >>    8.00     0.00   64.20  11.38   0.00
> >
> > Try setting the scheduler to 'deadline' and see if the queue sizes
> > shrink.
> >
> > No raid1? Besides adding redundancy, it can help with read
> > performance. I would probably put the mail on a raid 10 though if I
> > had 4 disks to do so.
> >
> > -Ross
> 
> > Like this:
> >
> > kernel /vmlinuz-2.6.9-78.0.8.ELsmp ro 
> root=/dev/VolGroup00/LogVol00 elevator=deadline
> >
> > And this change will be for System Wide.
> 
> Doing this among other things helped quite a lot.  My iostat -x %util
> dropped from ~ 60% to ~ 22% now.  Of course at same time I updated
> from dual-core 2Ghz CPU to quad-core 2.4Ghz.  Also swapped motherboard
> from one with an "Intel ICH9R" Southbridge to a "Intel ICH7R" since I
> heard CentOS 4.x did not have drivers in kernel for ICH9R.

That's why I asked what kind of Controler the board had on it in a previous
post to you and stated ram was not the suspect problem. IMO if you keeped
the dual core proc and just switched to ICH7 Board you would have saved
money. Your utilization rate would probly stayed the same or no higher than
%30. Just to keep things in balance you will probly want to try the cfq
schedular with a high user load so every thing gets it fair share in
time_wait. Some people will contradict that it's about making the users
happy. When access time for one user takes longer than another then the
complaints start coming in. I would like to know the Proc Utilization per
core or are you running it in Single Core?

JohnStanley