How to configure a JBOD disk array (PowerVault) for Performance

[JRoughgarden]

We have an application that is experiencing I/O contention,
particularly in tempdb but also in two other databases. The data is
stored on mirrored PowerVault 220's, each with 10 of 14 possible disks.
The PowerVaults are JBOD devices, not true SANs. The current config has
four separate groups of physical drives assigned to distinct logical
drives for log files, tempdb, and the two app dbs. This means, for
example, that tempdb resides on one mirrored drive. The standard advice
when faced with disk contention is to add spindles if possible. With 4
empty slots, we would presumably assign the new physical disks to the
most stressed db, e.g. tempdb.

An alternative arrangement would be to combine all the physical drives
into one logical drive, and put all the files, log and data, onto the
single logical drive. The hope for this configuration is that the
PowerVault would automagically distribute the data among the drives
such that all drives were in use, all spindles reading and writing at
maximum capacity when necessary. It is my understanding that
full-featured SANs, like NetApps and EMC models, do this. My question
is whether this configuration is best for the PowerVault, as well. Or
is this the essential difference between JBOD and a true SAN?

Has anyone tried both arrangements?

Advice is much appreciated.

[John Bell]

Hi

I can't say I know anything about the internals of JBOD.. but I would expect
combining the drives into a single logical drive to not be benificial even
though they are distributing the data over more spindles the type of access
between log file and data files will probably be counter productive.

If you are having contention on tempdb you may want to consider splitting
tempdb into multiple files (preferrably on separate sets of spindles) as
described in http://support.microsoft.com/default...b;en-us;328551

John

"JRoughgarden" <jr**********@stanfordalumni.org> wrote in message
news:11*********************@o13g2000cwo.googlegro ups.com...

We have an application that is experiencing I/O contention,
particularly in tempdb but also in two other databases. The data is
stored on mirrored PowerVault 220's, each with 10 of 14 possible disks.
The PowerVaults are JBOD devices, not true SANs. The current config has
four separate groups of physical drives assigned to distinct logical
drives for log files, tempdb, and the two app dbs. This means, for
example, that tempdb resides on one mirrored drive. The standard advice
when faced with disk contention is to add spindles if possible. With 4
empty slots, we would presumably assign the new physical disks to the
most stressed db, e.g. tempdb.

An alternative arrangement would be to combine all the physical drives
into one logical drive, and put all the files, log and data, onto the
single logical drive. The hope for this configuration is that the
PowerVault would automagically distribute the data among the drives
such that all drives were in use, all spindles reading and writing at
maximum capacity when necessary. It is my understanding that
full-featured SANs, like NetApps and EMC models, do this. My question
is whether this configuration is best for the PowerVault, as well. Or
is this the essential difference between JBOD and a true SAN?

Has anyone tried both arrangements?

Advice is much appreciated.

[JRoughgarden]

Yes, we have split tempdb in accordance with this article. No obvious
improvement yet. Looking at the -T1118 flag in a replicated
environment.

I am gathering that a JBOD is just a holder for physical disks that is
capable of splitting them up into logical disks, with or without RAID.
No real optimization.

Still, it would be an interesting experiment to put everything on one
logical disk comprising 14-physical disks vs. splitting the 14 into
separate logical drives for log files, tempdb mdf, and app data file
mdfs.