Is there anyone out there that knows of any performance data on using OS
striping over array hardware striping. What I mean is, has anyone
experienced or seen any documentation on the benefit of using an OS or
Veritas striping function over logical volumes that are made up of RAID luns
on a high end array.


Thanks in advance.


Gary

There's too many factors for specific performance data, but if you have
RAID hardware, its likely to do a lot better than software striping.
Better to let the firmware on the RAID unit handle it than the CPU and
SCSI bus on the system, which have enough other things to do.

Gary Jahrig wrote:
> Is there anyone out there that knows of any performance data on using OS
> striping over array hardware striping. What I mean is, has anyone
> experienced or seen any documentation on the benefit of using an OS or
> Veritas striping function over logical volumes that are made up of RAID luns
> on a high end array.
>
>
> Thanks in advance.
>
>
> Gary
>
>

"Gary Jahrig" <gary@X.usnetcastX.com> wrote in message news:<WZ2Ib.42397$m83.17913@fed1read01>...
> Is there anyone out there that knows of any performance data on using OS
> striping over array hardware striping. What I mean is, has anyone
> experienced or seen any documentation on the benefit of using an OS or
> Veritas striping function over logical volumes that are made up of RAID luns
> on a high end array.
>
>
> Thanks in advance.
>
>
> Gary

We did that with EMC boxes and LVM for a long time and saw better
performance than without the software striping - I don't know how and
I wasn't around in the days when it was being architected. Everyone at
EMC except for one guru said not to do it...

We are now using HDS (Hitachi - same as HP XPs) disk frames and no
longer doing software striping. Can't really compare the two because
the HDS frames are way faster anyway.

I could never understand the desire to do both, as it seemed intuitive
that the OS overhead would lessen the overall performance, but
according to the folks I work with, it did not.

Kevin

In comp.sys.hp.hpux Kevin Collins <spamtotrash@toomuchfiction.com> wrote:
> We did that with EMC boxes and LVM for a long time and saw better
> performance than without the software striping - I don't know how and
> I wasn't around in the days when it was being architected. Everyone at
> EMC except for one guru said not to do it...

> We are now using HDS (Hitachi - same as HP XPs) disk frames and no
> longer doing software striping. Can't really compare the two because
> the HDS frames are way faster anyway.

> I could never understand the desire to do both, as it seemed intuitive
> that the OS overhead would lessen the overall performance, but
> according to the folks I work with, it did not.

Using more volumes (bound as physical drives to the LVM) offers the OS more
than one disk queue. This can boost performance if you take care not to bind
more than one LVM volume to the same RAID-group.

urle

--
Uli Betzler (urle) urle@rz.uni-karlsruhe.de
Universitaet Karlsruhe, Rechenzentrum IRA/VERA/SR, Zirkel 2, 76128 Karlsruhe
phone: +49 721 608-4039 FAX: +49 721 608-9013 alternate FAX: +49 721 32550
urle = URL-Emanation = http://www.uni-karlsruhe.de/~Ulrich.Betzler/urle.html

Kevin Collins wrote:
> "Gary Jahrig" <gary@X.usnetcastX.com> wrote in message news:<WZ2Ib.42397$m83.17913@fed1read01>...
>
>>Is there anyone out there that knows of any performance data on using OS
>>striping over array hardware striping. What I mean is, has anyone
>>experienced or seen any documentation on the benefit of using an OS or
>>Veritas striping function over logical volumes that are made up of RAID luns
>>on a high end array.
>>
>>
>>Thanks in advance.
>>
>>
>>Gary
>
>
> We did that with EMC boxes and LVM for a long time and saw better
> performance than without the software striping - I don't know how and
> I wasn't around in the days when it was being architected. Everyone at
> EMC except for one guru said not to do it...
>
> We are now using HDS (Hitachi - same as HP XPs) disk frames and no
> longer doing software striping. Can't really compare the two because
> the HDS frames are way faster anyway.
>
> I could never understand the desire to do both, as it seemed intuitive
> that the OS overhead would lessen the overall performance, but
> according to the folks I work with, it did not.
>
> Kevin
We have used internals/Jamaca boxes for software and MTI external arrays
for hardware and have always gotten better perf out of the hardware.
With hardware RAID the OS doesn't have to do anything, it does make a
diff. It also depends on what interfaces you are using though.

> Using more volumes (bound as physical drives to the LVM) offers the OS
more
> than one disk queue. This can boost performance if you take care not to
bind
> more than one LVM volume to the same RAID-group.
>
It also helps balance the load over multiple host interfaces, if you have
them.