This issue of mixing the number of disks in the raid groups is covered in the MetaLUN White Paper found in the Documents section on this forum

White Paper EMC CLARiiON MetaLUNs - A Detailed Review.pdf

There is no issue with the number of component LUNs that you can use for a metaLUN - just that four component LUNs when using 4 4+1 Raid 5 riad groups has a natural alignment.

glen

You won't have an optimal configuration for performance, but you can create a striped or concatenated metalun with those two LUNs and it will work fine.

The recommendation you quoted is for the best performance configuration. So if you aren't trying to create the most optimal configuration from a performance perspective then don't worry about it. I have deployed striped metaLUNs built from only two LUNs and have never had any issues with them. It was not an ideal configuration but it meet the requirements of the project which did not have performance at the top of the list.

Like AranH said, if your concerned with performance, then that setup is not optimal. The reason would be disk count wise. The first group (number of drives in the group) helps to determine how much data is placed on each striped metalun component before moving to the next component. This in your config will cause the data to be more compacted on the smaller setup (less drives to put the data on) compared to the second group. If you start hitting the lun hard, the performance decrease will start with the smaller group due to the number of drives in the group. If not a performance concern, then you should be ok.

I tried to sum it up, but here is the MetaLun Bible for more information: White Paper: EMC CLARiiON MetaLUNs — A Detailed Review

-Ryan

Thank you all!!

It seems to come right down to performance expectations with most of the supported config's.

Allow me to ask one more thing here...

If I create a 4 LUN Meta living on 3 DAEs with 2 LUNs on 1 DAE (in adjacent RGs) is this also 'merely' a performance issue or something else?

Thanks again!!

Performance degradation doesn't cross raid groups on a bus unless there is a fault that impacts the groups or you exceed the physical capabilities of the bus (a 2Gb or 4Gb connection). I think exceeding the capabilities of the bus is pretty rare. If 1 group is having a perfomrance issue due to its' layout, another raid group next to it shouldn't be impacted. I don't see any issues with creating 4 luns across 4 raid groups on the same bus and striping them together. Just want to make sure if you are striping to make sure the raid groups have the same number of drives in each.

-Ryan

Thanks Ryan!!

My delima here is coming into an environment that was setup and carved out by someone who did not know what they were doing.

Essentially just throwing space at servers that needed it.

Although the RAID type is consistent through the config, R5, the number of disk is not.

There are 4+1, 5+1 & 6+1's and they are all used is various different 2 LUN METAs.

If I am understanding you right, I should be concerned with the number of disks in each component of the META?

I may have misread a previous post but it sounded like this was not an issue (just not optimal)...

Thanks!!

Thanks Aran.

I've read through the pertinent areas in the MetaLUN doc. I was just looking for clarification from what sounded like a contradictory response.

There is no substitute for experience, and most of the answers I'm getting are (paraphrasing) 'you can meta whatever', 'performance may be impacted' and such. Which is great, as there is no going back at this point.

Thank you all, your answers and links have been quite helpful!

You're welcome, glad I could help.

There are lot of things we "can" do with CLARiiONs that are not ideal but work and provide the storage needed for our environments. As storage endusers we often don't have the budget, time, or ability (due to inheriting misconfigured systems) to implement the best practices for every host or array. The guides are helpful enough in telling us what we can do but the comments you will see from the EMCers will mostly be in support of the best practices and warning you about the result of not following those practices.

The MetaLUN guide that Glen posted a link to states that you can create striped metas using component LUNs with different disk counts. It states that this is not ideal though from a performance perspective.

Distributing your LUNs across more than one DAE may have a positive effect, but may not have any effect at all.  For example, you'd have to be certain the DAEs were on different busses.  Your primary concern should be that any LUNs that are sharing your MetaLUN's component LUNs RAID group has complementary I/O.  This avoids disk contention.

A detailed discussion of distributing LUNs across DAEs is found in the 'Bus Balancing' section of the EMC CLARiiON Best Practices for Performance and Availability, FLARE Revision 29.0. A discussion of Complimentary I/O is found in the previously recommended 'Partitioned RAID Groups' section of the EMC CLARiiON MetaLUNs paper. Both documents are available on Powerlink.