RE: [suse-sles-e] File systems and SCSI Arrays...

From: Anders Norrbring (lists_at_norrbring.se)
Date: Wed Feb 16 2005 - 18:19:45 CET


From: "Anders Norrbring" <lists@norrbring.se>
Date: Wed, 16 Feb 2005 18:19:45 +0100
Message-Id: <20050216171945.35897D91B@mail.the-server.net>
Subject: RE: [suse-sles-e] File systems and SCSI Arrays...


> On Wednesday, February 16, 2005 08:55 am, Anders Norrbring wrote:
> > > On Wednesday, February 16, 2005 03:28 am, Anders Norrbring wrote:
> > > > > On Tuesday, February 15, 2005 01:28 pm, Anders Norrbring wrote:
> > > > > > The disk space is located on a Compaq Smart Array 3200
> > > > > > controller. The controller takes care of everything regarding
> > > > > > RAID, so that part doesn't load the Linux system.
> > > > > > The system loads the driver cpqarray in initrd, everything
> > > > > > seems fine so far.
> > > > > >
> > > > > > The problem is that when the file system gets a load, for
> > > > > > example if I copy large files to the volume, it takes forever
> > > > > > to complete and the system seems hanged.  After 10-90
> > > > > > seconds, it continues its work. If I have one window open
> > > > > > with 'top', that too freezes during this period.
> > > > > >
> > > > > > Since it's SuSE SLES9, I formatted everything with reiserfs,
> > > > > > my thinking wass that this may be the problem, but then I
> > > > > > tried with XFS as well, just the same problem..  I'm just
> > > > > > guessing..  If someone can clear things up, I'd be grateful..
> > > > >
> > > > > Anders,
> > > > >
> > > > > I've got a SLES9 box running on an old Compaq 6400r with a
> > > > > SmartArray 3200 controller and I see none of the issues you are
> > > > > having. Reiser shouldn't make a difference (although I now
> > > > > prefer ext3...)
> > > > >
> > > > > If your box is a Compaq box, did you SmartStart it before
> > > > > installing SuSE? (The choice of OS to be installed sets the
> > > > > BIOS up correctly.)
> > > > >
> > > > > Also, check your boot options in grub and look for the
> > > > > "elevator= " parameter. If it is set to "as", try changing it
> > > > > to "cfq" or "deadline" and see what happens after a reboot.
> > > > > (My Compaq has the parm set to "cfq"; as I understand it, RAID
> > > > > controllers will see poorer performance if the parm is set to
> > > > > "as".)
> > > >
> > > > Thanks Mark,
> > > >
> > > > No, this isn't in a Compaq system, so I can't use the SmartStart
> > > > CD.. In my grub's menu.lst, I have:
> > > >
> > > > kernel (hd0,1)/boot/vmlinuz root=4802 vga=0x31a splash=silent
> > > > resume=/dev/ida/c0d0p1 showopts
> > > >
> > > > So, there's no "elevator" at all set, perhaps that's the
> > > > problem... I really haven't spent any time to study kernel and
> > > > boot options, since I haven't found any documentation at all
> > > > regarding this old SA 3200 controller. I've just moved it from an
> > > > old Compaq server that ran W2k..
> > > >
> > > > I guess I need to find boot/kernel options that can have effect
> > > > on the Smart Array, I just don't know where to start... I'll try
> > > > to set "elevator=cfg" in my grub as well, perhaps it can do some
> > > > good.. :)
> > > >
> > > > Later,
> > > > Anders Norrbring
> > > > Norrbring Consulting
> > >
> > > Anders,
> > >
> > > Here's the boot line from my Compaq server's SLES9 kernel grub
> > > setup. Note that the elevator line is "cfq" not "cfg"! Our config
> > > lines look verrry similar, but YMMV...
> > >
> > > Best regards,
> > > Mark
> > >
> > > (hd0,1)/boot/vmlinuz root=4802 vga=0x317 selinux=0 splash=silent
> > > resume=/dev/ida/c0d0p1 elevator=cfq showopts
> >
> > I looked in the kernel's Documentation directory and found out when
> > reading cpqarray.txt and as-iosched.txt that "elevator=deadline"
> > would be a good choice, and so it seems.. :)
> > Now the system looks both stable and fast, I'll monitor it and
> > provoke it with some huge file copying..
> >
> > Anders Norrbring
>
> I use elevator=deadline on my personal workstation, which has a DPT
> 2865U3 caching RAID controller with a 4-disk RAID5 array attached. The
> performance, especially reads, is much improved over the SUSE default
> setting of "elevator=as".
>
> Before you put your server into production, might I ask if you would be
> interested in testing the box with both "elevator=deadline" and
> "elevator=cfq", and posting your results?
>
> If elevator=deadline gets you better performance without sacrificing
> stability, I'd be keen to use that setting. But since my
> 3200-controller-equipped server performs some very critical functions
> 24x7x365, I'm hesitant to change the setting from the SuSE default
> without some confirmation from the "real world" on identical hardware.

It seems like I was wrong.. I just made a copy of 200MB data in 6 files from
an IDE drive to the RAID. After just 50MB or so, the IDE activity went to
zero, while all the drive LEDS on the array are flashing like mad. Also,
the console appears to be "dead" or locked up during extended amounts of
time, between 5 and 25 seconds at a time.

After that, I can use console a little while, and then the lockup comes on
again.

My Smart Array 3200 is set to 50/50 read/write caching, so it shouldn't be
as slow as the system seems to think it is.

My best bet is that the cpqarray driver isn't totally healthy, but I'll try
to set elevator=cfq before I go on complaining... I doubt the controller is
not fully functional, it races in Windows.. I could put in another one,
just to try.

Anders Norrbring

---------------------------------------------------------------------
To unsubscribe, e-mail: suse-sles-e-unsubscribe@suse.com
For additional commands, e-mail: suse-sles-e-help@suse.com



This archive was generated by hypermail 2.1.7 : Wed Feb 16 2005 - 18:21:15 CET