Disk Enclosures
1748233 Members
3259 Online
108759 Solutions
New Discussion юеВ

Re: Performance XP24K

 
Juan Carlos Mendoza Cal
Occasional Contributor

Performance XP24K

Hi, I have a XP24K I need to know what is the max performance in IOPS for each disk 146GB 15K RPM, what is the max for Array group 7D+1P in groups the 4 concatenation. IOPS for each MP the CHIPs, and ACPs.

Other question, i need documentation for Performance in XP24K, manuals, tips, etc. and best practices
7 REPLIES 7
Juan Carlos Mendoza Cal
Occasional Contributor

Re: Performance XP24K

Other question, what is emulation for best performance Open-9 and create LUSE or OPEN-V, and what is the diferences
Amar_Joshi
Honored Contributor

Re: Performance XP24K

Juan,

For the first question, the answer can be very long and need opinions and experience to get till there. I have attached the ESG's white paper, read on from page 16 and you will learn how the configuration tricks the performance. Some of the numbers can be found from HP's website where they are talking what is the difference between XP12k and 24k.

For the second question, Open-V is always preferred over other emulations, reason that Open-V uses less of internal configuration memory space to build the volumes and second is that LUSEs degrades the overall LDEV performance. So there is a plus for Open-V and a negative for LUSE that will give you a ++ to go for Open-V.

Hope that helps.
IBaltay
Honored Contributor

Re: Performance XP24K

Hi,
Some terms need to be explain first:
Parity group (PG) -> physical group of disk on one loop (it is 8 disks -> 7D+1P, 6D+2P, 4D+4P). The PGs can be concatenated into max 28D+4P. Therefore from the perf reasons the best approach is if the host LUN is distributed across many PGs and then LVM on the host side:
example:
lets assume that you want to have a 100GB LUN from XP(24000) for Unix/Linux host.
a)Then it is much better to create 4x25GB or 5x20GB (RAID5) and give it to the host LVM and even LVM stripe them.
b) if we create one "big" 100GB LDEV then we are limited by the number of disk in the PG and 8 disks - that is only few spindles
c) if we create the same as in a) but as LUSE, then yes we are doing the distribution across the PGs, but unfortunately LUSE is only spanned not stripped, that means that all writes will go first to the LDEV nr. 1 only, then LDEV nr. 2, etc.

maybe the question is the LDEV or LUSE volume. The main difference that LDEV is striped and LDEV is not (only spanned volume).
The par
the pain is one part of the reality
IBaltay
Honored Contributor

Re: Performance XP24K

Sorry, i am resending this once more with the corrected end :-):

Some terms need to be explain first:
Parity group (PG) -> physical group of disk on one loop (it is 8 disks -> 7D+1P, 6D+2P, 4D+4P). The PGs can be concatenated into max 28D+4P. Therefore from the perf reasons the best approach is if the host LUN is distributed across many PGs and then LVM on the host side:
example:
lets assume that you want to have a 100GB LUN from XP(24000) for Unix/Linux host.
a)Then it is much better to create 4x25GB or 5x20GB (RAID5) and give it to the host LVM and even LVM stripe them.
b) if we create one "big" 100GB LDEV then we are limited by the number of disk in the PG and 8 disks - that is only few spindles
c) if we create the same as in a) but as LUSE, then yes we are doing the distribution across the PGs, but unfortunately LUSE is only spanned not stripped, that means that all writes will go first to the LDEV nr. 1 only, then LDEV nr. 2, etc.

Resume:
Unix/Linux -> smaller LDEVs and LVM
Windows -> there are some FS (NTFS) and therefore LVM limitations. So here the PG concatenation helps - max is 32 disks (28D+4P) and LUSE should be distributed across the different PGs.


the pain is one part of the reality
IBaltay
Honored Contributor

Re: Performance XP24K

Hi,
this thread can also help:
http://forums11.itrc.hp.com/service/forums/questionanswer.do?admit=109447626+1217957643931+28353475&threadId=984111

and here is the xp12000 best practices, (yes it is xp12000 but the main recommendations are the same)
http://h71028.www7.hp.com/ERC/downloads/4AA0-7923ENW.pdf
the pain is one part of the reality
Juan Carlos Mendoza Cal
Occasional Contributor

Re: Performance XP24K

The document, is bad only see pictures, but the text don├В┬┤t can see.
IBaltay
Honored Contributor

Re: Performance XP24K

Hi,
i am downloading it to you to see if there is any difference, because "mine" works :-)
the pain is one part of the reality