RE: Linux performance
Posted in 2000
Agree with Art. Amount of cache on Raid Controller, Writeback or
writetrough, Stripe size, number of spindles, SCSI ULTRA #, PCI Bus speed,
IO Chip/Mhz {Most use to have Intel 960} all have a big influence on
performance.
Also look at Megaraid controllers. Last tested beat Mylex and easier to set
up. This was 3 years ago and for sure could have changed by now.
DTP + Compaq also high perf controllers.
Adaptec cheaper level that will suite Linux.
-----Original Message-----
From: Art S. Kagel [mailto:kagel@bloomberg.net]
Sent: Wednesday, 20 December, 2000 1:18 AM
To: informix-list@iiug.org
Subject: Re: Linux performance
John, are you using Linux RAID? It beats hell out of the CPUS! Faster than
a speeding bullet but it gets that speed by sucking up all the CPU cycles it
can get. Linux built in software RAID is only really good for configuring
a dedicated fileserver not for any machine that needs the CPU also. For
serious RAID get a hardware RAID10 solution like ICP Vortex or Mylex. Those
controllers have their own CPUs and offload all the RAID work to them
leaving
the main system CPUs alone.
In a recently published test, between Linux software RAID5 and Vortex RAID5
on a Linux box, Linux gave better throughput but maxed the two test machine
CPUs out doing it.
Art S. Kagel
Obnoxio The Clown wrote:
>
> In the year of Our Lord Tue, 12 Dec 2000 14:50:53 -0700, John Miller
> <jmiller@itc.nrcs.usda.gov> broke a vow of silence to utter:
>
> >
> >We have a 4-700 Mhz cpu Intel box with the 2.2.16 linux Kernel(SuSE
> >7.0).
> >We have the number of CPU's set to 3 for Informix, 1 less then the max.
> >The problem is once we get more then 4 users hitting the database hard
> >the system starts to crawl. iostat shows the CPU's running at 80 to
> >100%.
> >Does anyone know if this is the best performance we can get?
> >
> >We are using RAID 1+0 on two controllers and raw space.
>
> Bloody hell!! How is the CPU load spread? Mostly user, or lots of sys or
wio?
>
> Can you post output for:
>
> onstat -p
> onstat -P | tail -5
> onstat -R
> onstat -F
> onstat -g seg
> onstat -d
> onstat -D
> onstat -m
> onstat -c> sar 5 12