Merits of 1 massive EVA disk vs several small

[Ian Dennison_1]

OK, I have consultants (non-HP) here putting forward the merits of 1 massize disk on the EVA per database (the standard windoze philosophy). I have always touted the "several smaller disks" (8-16 per VG, each up to 50GB in size) as being better for HP-UX and for SAN-connections / HBAs in between (multiple paths, load sharing)

I have searched the forums and cannot find any discussions that address the merits and drawbacks of each method. Am I right in concluding that several disks allow HP-UX to manage disk queues better (with no impact on performance)?

Share and Enjoy! Ian Dennison

[Naveej.K.A · Accepted solution]

Hi Ian,

Have a look at this link.
Hope this helps..

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=586590

with best wishes
Naveej

[Jean-Luc Oudart]

Also these ones
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=254146

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=550429

Regards,
Jean-Luc

[Ashwani Kashyap]

Yes you are right . Not only are you spreading your I/O nicely by having several small LUN's but also since HPUX has very small I/O queue length per LUN ( Ithink about 8 ) you are effectivley increasing the queue to accept more I/O requests .

[Ian Dennison_1]

Thanks fellow forummers for your feedback so far.

Interesting question: if you have 1 disk queue dealing with many disks requests (and queueing), is this faster or slower than many disks with their own queues.

Can HP-UX only work on 1 disk request at a time? Is one long disk queue just as manageable / efficient as several short disk queues?

Look forward to your thoughts. Share and Enjoy! Ian

[Ashwani Kashyap]

The serice time for any I/O request will certainly go down drastically if you have many queues instead of one , meaning i/o's will be faster .

[Uwe Zessin]

The EVA will spread the I/Os over all physical disks in the disk group anyway through its DVR (distributed virtual RAID) technlogy. I thought I have read the other day that one can increase the queue length on HP-UX.

HP-UX can, of course, work with multiple I/O requests at a time. The queue length specifies the maximum number of 'in flight' I/O operations. A value too high will overrun the disk array - the array can tell the server to throttle back, but it depends on the quality of the drivers in the operating system if the server obeys.

A single I/O queue allows the array to better recognize certain I/O patterns like sequential I/O, so it can do read-aheads. That won't work with host-based striping.