>> Would it be better to stripe at the storage level (EMC) or at the O/S
>> level (Solaris 2.6/2.8)? The reason I ask is that if striping is done
>> at the EMC level, a single striped device (LUN) is presented to the
>> O/S. If striping at the O/S, EMC will present multiple LUNs to the
>> O/S which will then be striped. Are there any bottlenecks introduced
>> by not making the O/S aware of the actual devices that are being
>> striped across? In other words, even though the single striped device
>> presented by EMC would be striped, do we lose any performance since it
>> a single I/O device being presented to the O/S?
> There's no point in doing software striping if you can do it in
> hardware. If you wanted softwer striping, why giving $$$$$$ for
> something like EMC? Do striping in hardware and use CPU cycles on
> your servers for something more usefull.
Depends on lots of things. For instance, if you're using SCSI, you
could saturate a channel. If you stripe at the OS level across the
channels, then you get an I/O win.
Even with faster channels like FC-AL and Fabric FC, you might still get
a boost.
The OS has to do all the work to figure out which disk gets the I/O. It
is not likely to be much more work to do a stripe.
Some configurations of EMC may limit your future flexibility. Can you
change the stripe size later without taking your data offline?
--
Unix System Administrator Taos - The SysAdmin Company
Got some Dr Pepper? San Francisco, CA bay area
< How are you gentlemen!! Take off every '.SIG'!! >
by