This post is more than 5 years old
108 Posts
0
1109
Symmetrix Meta Volume Performance
We have a HPUX Cluster that does have around 100 Luns of 21GB from HDS array and this is going to be migrated to DMX. So I need to provide the LUNS from DMX.
What is the best preferred meta size for this? I have device size of 11.8GB on DMX. Is it better to go ahead with 100 * 23GB devices? or increase the device size to something like 71GB?
Are there any performance constraints ? Can some one provide some idea on how does the device affect?
What are the things that i need to consider?
What is the best preferred meta size for this? I have device size of 11.8GB on DMX. Is it better to go ahead with 100 * 23GB devices? or increase the device size to something like 71GB?
Are there any performance constraints ? Can some one provide some idea on how does the device affect?
What are the things that i need to consider?
xe2sdc
2 Intern
2 Intern
•
2.8K Posts
0
August 19th, 2008 08:00
1) with hpux you have a powerful volume manager .. you don't need to have deivces with same size as before .. Yes I'm talking about LVM
2) think at your 11.8 Gb devices as building blocks .. Now use metavolumes to build your Lego city .. It's suggested to have an even number of members when you form your metavolumes. If you are already using RAID5, form concatenated metas, if you have 2-way mirror devices, form striped metas.
3) Having a lower number of bigger luns usually have little if no impact .. However reducing number of drives usually puts more pressure on queues at host and storage level (since less luns means more iops for every lun). Usually you don't have problems at storage level (since your cache depends on number of metamembers, however you form them in metas). YMMV
My 2 cents ...
SKT2
2 Intern
2 Intern
•
1.3K Posts
0
August 19th, 2008 12:00
is that normal cluster or with RAC extention ?
AJK3
108 Posts
0
August 19th, 2008 12:00
So Can i take this as that using metas of size 23GB is better than metas of 71GB?
any other points that can be discussed or can be looked at?
xe2sdc
2 Intern
2 Intern
•
2.8K Posts
0
August 19th, 2008 15:00
Generally speaking bigger metavolumes are better from storage perspective.. however it depends on actual hardware configuration (number of drives/DA pairs) and protection (raid5 VS mirror).
At the same time bigger metavolumes means less queues in the host so it's worst from host perspective. Since you already have 21Gb devices and they give you good performances, using 23Gb metavolumes may be fine too. Eventually considere doubling size.
But don't go too far since bigger metavolumes means also bigger chunks of space you give to your host when you expand capacity. If you have 23Gb metavolumes you can grow from 138 to 161 Gb ... If you have 69Gb metavolumes, you go straight from 136 to 207 Gb
Message was edited by:
Stefano Del Corno
added thread name and comments
wishtobeMrEMC
155 Posts
0
August 20th, 2008 00:00
Really appreciate if you explain technially on the above statement. Actually, i had a case that unix admin prefers to have smaller luns [raid 5]for oracle db with the assumption that more luns means more disks thus the more spindles for iops.
xe2sdc
2 Intern
2 Intern
•
2.8K Posts
0
August 20th, 2008 10:00
It depends on a lot of factors .. however having everything else fixed, an 8way metavolume usually performs better then a 4way metavolume .. with "everything else fixed" I mean same size of meta members, same number of drives, same number of da-pairs, same cache, same protection same-everything-you-may-think-about. If you use bigger metavolumes you'll spread your workload on a bigger number of HDA in the backend.
You said you are using RAID5 .. RAID5 already uses more spindles .. And when you form metavolumes you should use members from different "raid groups" (even if you don't have real raid groups in DMX).
AJK3
108 Posts
0
August 20th, 2008 11:00
xe2sdc
2 Intern
2 Intern
•
2.8K Posts
0
August 20th, 2008 12:00
RRR
2 Intern
2 Intern
•
5.7K Posts
0
August 31st, 2008 12:00