Slow SCSI Disk Access on AMI Elite 1600 controller

Slow SCSI Disk Access on AMI Elite 1600 controller

[Ryan Shrout]

Okay, I am fairly new to all this, so bear with me as I try to explain the
problem fully.  :)

First, my problem was centered around Mysql.  I had a problem of processes
spawning and spawning (sometimes over 100 of them at a time).  A simply
mysqld restart would solve the matter for most of the day, but the problem
always persisited.  One the people in the mysql list pointed me to check my
disk performance.  I came up with this:

[root@dagger /root]# hdparm -Tt /dev/sda

/dev/sda:
 Timing buffer-cache reads:   128 MB in  0.93 seconds =137.63 MB/sec
 Timing buffered disk reads:  64 MB in 13.92 seconds =  4.60 MB/sec

My systems specs are:
Tyan Thunder K7 Dual-Athlon (non-Via chipsets) motherboard
2 - 1.2 Ghz Athlon MP CPUs
1 GB PC2100 Registered DDR DRAM
AMI Elite 1600 SCSI RAID controller with 64 MB SDRAM
4 - 15K RPM Cheetah SCSI HDDs
Red Hat Linux 7.2

Obivously, 4.60 MB/sec is WAY too slow for this setup, and the mysql people
assumed this was probably what was causing the Mysqld server to crap out.

So, I then started trying to figure out to raise the buffered disk read
speed.  My only solution I came across -- find out if the SCSI
controllers/drives were in asynchronous mode and if they are, change them to
synchronous mode.

Now, how can I tell what mode my SCSI disks are in and how can I change it
to synchronous if it isn't set that way already?

Thank you!!!

Ryan Shrout
Owner - Amdmb.com
http://www.amdmb.com/
rshrout@amdmb.com

Re: Slow SCSI Disk Access on AMI Elite 1600 controller

[Ryan Shrout]

> >  Timing buffered disk reads:  64 MB in 13.92 seconds =  4.60 MB/sec
> ...
> > 4 - 15K RPM Cheetah SCSI HDDs
>
> each disks will sustain 30-50 MB/s, so yeah, that's kind of low.
>
I changed the drives settings from Write-Through to Write-Back and it went
to 22.5 MB/sec

> > AMI Elite 1600 SCSI RAID controller with 64 MB SDRAM
>
> I wish people would start with relatively simple controllers,
> and only move to fancy ones if they know they need them,
> and can prove that the fancy controller works better.
>
> so for instance, these disks on any PCI ultra2 controller
> will easily sustain 100 MB/s.  SW raid5 might slow things down
> if you had, say, a single celeron (260 MB's dram bandwidth),
> but certainly wouldn't effect a dual athlon with ddr.
>
I am using them in a RAID 0+1 configuration, so this kind of card was
needed.  Besides, it was a gift (Free!) and I couldn't pass that up.

> > Red Hat Linux 7.2
>
> do you mean with RH's kernel!?!
>
No, actually, its 2.4.6-2smp custom compiled.

> > So, I then started trying to figure out to raise the buffered disk read
> > speed.
>
> do you have a simple, non-raid ultra2 controller around?
>
> > My only solution I came across -- find out if the SCSI
> > controllers/drives were in asynchronous mode and if they are, change
them to
> > synchronous mode.
>
> unless something dramatic is happening, the disks will stay in the
> same mode as they're detected at driver-init time, which should be
> right before your eyes in /var/log/dmesg.  well, at least for a non-fancy
> controller.  for a gold-plated controller like you have, you need
> to find some way to query the controller, perhaps a serial port,
> or some too provided by AMI.  basically that's a non-linux question,
> since the controller has its own little OS that you need to interact with.
>
> > Now, how can I tell what mode my SCSI disks are in and how can I change
it
> > to synchronous if it isn't set that way already?
>
> unless your scsi cabling is utterly botched, they'll be synchronous.
>
Okay, that helps me so I stop looking for that setting!  :)

Hopefully, going from 4.5 -> 22.5 MB/sec will be enough to temporarily stop
my mysqld crashing.

Thanks!

Ryan Shrout

Slow SCSI Disk Access on AMI Elite 1600 controller

[Ryan Shrout]

Okay, I am fairly new to all this, so bear with me as I try to explain the
problem fully.  :)

First, my problem was centered around Mysql.  I had a problem of processes
spawning and spawning (sometimes over 100 of them at a time).  A simply
mysqld restart would solve the matter for most of the day, but the problem
always persisited.  One the people in the mysql list pointed me to check my
disk performance.  I came up with this:

[root@dagger /root]# hdparm -Tt /dev/sda

/dev/sda:
 Timing buffer-cache reads:   128 MB in  0.93 seconds =137.63 MB/sec
 Timing buffered disk reads:  64 MB in 13.92 seconds =  4.60 MB/sec

My systems specs are:
Tyan Thunder K7 Dual-Athlon (non-Via chipsets) motherboard
2 - 1.2 Ghz Athlon MP CPUs
1 GB PC2100 Registered DDR DRAM
AMI Elite 1600 SCSI RAID controller with 64 MB SDRAM
4 - 15K RPM Cheetah SCSI HDDs
Red Hat Linux 7.2

Obivously, 4.60 MB/sec is WAY too slow for this setup, and the mysql people
assumed this was probably what was causing the Mysqld server to crap out.

So, I then started trying to figure out to raise the buffered disk read
speed.  My only solution I came across -- find out if the SCSI
controllers/drives were in asynchronous mode and if they are, change them to
synchronous mode.

Now, how can I tell what mode my SCSI disks are in and how can I change it
to synchronous if it isn't set that way already?

Thank you!!!

Ryan Shrout
Owner - Amdmb.com
http://www.amdmb.com/
rshrout@amdmb.com

Slow SCSI Disk Access on AMI Elite 1600 controller

[Ryan Shrout]

Okay, I am fairly new to all this, so bear with me as I try to explain the
problem fully.  :)

First, my problem was centered around Mysql.  I had a problem of processes
spawning and spawning (sometimes over 100 of them at a time).  A simply
mysqld restart would solve the matter for most of the day, but the problem
always persisited.  One the people in the mysql list pointed me to check my
disk performance.  I came up with this:



Ryan Shrout
Owner - Amdmb.com
http://www.amdmb.com/
rshrout@amdmb.com