Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

1603

May 10th, 2007 07:00

If a little is good, is more better? MetaLUNs

I have a server administrator requesting a MetLUN for his server. He knows a MetaLUN can only be stripe expanded in sizes matching the original size of the base LUN.

He wants a 250GB LUN in 10 25GB LUNs so if he needs to expand it he can do it in 25GB chunks.

He knows I can do it with two 125GB LUNs but he says anytime we want to grow it, we'll have to use another 125GB and he's not sure he's always going to need that much and might waste some of the space.

So my question is:

Does the number of components in a MetaLUN have ANY effect, good or bad on the performance of the MetaLUN?

I'm concerned about any negative performance impacts and I'd rather not make ten LUNs if I can do it with two!

76 Posts

May 10th, 2007 07:00

You won't actually have more than one component unless you add by means of concatenation, in which case you can add any size lun (in R24 anyway)
This presents the extension as a separate component, and does not stripe the data across the previous luns in the Meta.

If you choose to add further stripe extensions after the new component was created, the data will be striped across the newest component and any luns within it.

Keep in mind, if your going to stripe, the more luns that are striped together in a meta, the longer it will take to continue to expand more luns. Depending on the speed you select, this may provide a performance degradation

410 Posts

May 10th, 2007 07:00

but there would be a limit on max components you can put in a single metalun...16 i guess...not sure

76 Posts

May 10th, 2007 07:00

I should point out that if you extend a lun using concatenation, the lun that is added are added immediately without performance impact, but if another lun is stripped onto it, then data is restriped and performance is impacted.

410 Posts

May 10th, 2007 07:00

if you are going to stripe it, and each component comes from a different raid group, then more number of components will mean better performance.

23 Posts

May 10th, 2007 11:00

if you are going to stripe it, and each component
comes from a different raid group, then more number
of components will mean better performance.


It is going to come from the same raid group.
Yes, I am striping it.

23 Posts

May 10th, 2007 12:00

I am striping.

I am not concerned with how long it takes to build it. I am asking if the number of LUNs in a MetaLUN affects the overall performance of the MetaLUN at all.

Thanks for your reply,

Pat


You won't actually have more than one component
unless you add by means of concatenation, in which
case you can add any size lun (in R24 anyway)
This presents the extension as a separate component,
and does not stripe the data across the previous luns
in the Meta.

If you choose to add further stripe extensions after
the new component was created, the data will be
striped across the newest component and any luns
within it.

Keep in mind, if your going to stripe, the more luns
that are striped together in a meta, the longer it
will take to continue to expand more luns. Depending
on the speed you select, this may provide a
performance degradation

23 Posts

May 10th, 2007 12:00

I am striping.
I am not concerned with how long it takes to create the metaLUN.

The question was:

Does the number of component LUNs in a metaLUN have any affect on it's overall performance after it goes into production?

2.2K Posts

May 10th, 2007 14:00

ppage,
As Kiran pointed out, the performance benefits that come from the use of MetaLUNs occur when LUNs from different RAID groups are stripped together.

You said that you would be using LUNs all from the same RAID group? I don't know what benefit this provides you other than being able to grow the LUN as the server administrator wants, but I don't recommend a storage architecture based on what a "server administrator" wants. Better off just making the RAID group a dedicated RAID group (one LUN spanning the entire RAID group) for that application. If you are assigning other LUNs from that RAID group to other servers/applications then that will produce more of a decrease in performance more than the striping of multiple LUNs within the same RAID group.

I don't think it would degrade performance, but I have never seen metaLUNs created using all LUNs from the same RAID group.

Aran

410 Posts

May 10th, 2007 16:00

the reason to use different RAID group is to avoid elevator seeks on I/Os.

Assuming there is no other I/O going on the RAID Group and a smallest I/O write comes on. The I/O will be completed by just one disk head movement and write cycle on each drive of the RAID group.
Now if you use another LUN from the same RAID group, there would be two write cycles on each drive and a head seek in between. And if there is a fair amount of sequential write on the RAID group, the disk head will move more times than it would have if you had used different RAID Groups.

CLARiiON cache tries to smooth out burst of I/Os on the disks but if the writes are dependent, then its not much of use. So its better to stick to different RAID groups with striped metaluns

2.2K Posts

May 11th, 2007 11:00

Thanks for the clarification Kiran. Good answer. So it sounds to me like you answered ppage's question. All in all not a good idea to create a metaLUN using LUNs from the same RAID group.

410 Posts

May 14th, 2007 03:00

thanks Aran,
above para applies to striped metalun.
with concatenated metas, I/Os go on to one component at a time and there we can combine luns from same RAID group.
No Events found!

Top