AFAIK there's no such limit at 500GB. I was told in a clariion performance workshop that for example on ATA RaidGroups you have to create a single LUN on a RaidGroup and with ATA disks nowadays being 500GB large, this makes 1 LUN on a 8+1 RAID5 RG as large as 4TB (8x500).
Based on this I assume that on FC disks this is the same.

BUT: meta luns perform better, because they are layed out on more than 1 RG and thus invoke more physical spindels.

With the data you gave me, I would say that creating a meta lun made out of 4 luns ON DIFFERENT RG's is probably a good idea. DON'T use luns on the same RG as this will seriously impact performance on this RG.

A single R5(14+1) raid group with 2 TB LUNs carved out of would would provide much worse performance than 3 x R5(4+1) raid groups with 2 TB meta LUNs carved out, right?

BTW, EMC didn't say you couldn't create LUNs larger than 500 GB's, they said that performance suffered greatly.

Let's say I have a DAE full with 15 x 1 TB SATA disks and it will be used in a VMware environment. I don't require ton's of performance, but I do want to configure it for the best performance I can get.

The largest VMFS I can create without using extents is 2 TB's. So assuming I want to create 2 TB meta LUNs, what is the best way to configure?

That might have been for an old version of FLARE, there are no such restrictions these days. Like RRR said I can create a 500GB LUN using R5(2+1) on 500GB disks and the performance would suck compared to let's say a 1TB LUN created on a R5(8+1) using 146GB disks.

If you have one shelf of 1TB disks and want to limit the size of the LUN created using those disks but also get decent performance one possible option is to create 3 R5(4+1) RAID Groups and carve each RG into equal size slices and create metaluns using a slice from each RG. That way you have metaluns that are spread across 15 spindles that are also large in size. Of course that doesn't leave room for a 1TB hot spare...

Yes. The EMC document CLARiiON Best Practices for Fibre Channel Storage states that RAID Groups of more than ten drives results in decreased performance. You achieve better performance by using metaluns across the same number of spindles. The document is a good read, look for it in the white paper section on powerlink. You will see the EMC guys in the forum recommend it frequently for good reason: a lot of the questions asked in the forums are answered in it.

IMHO is performance better when you try to avoid having random I/O. As little as possible number of LUN's (preferably only 1) will do that. And if you should decide on making >1 LUN on a single RG: at leat give those LUN's to the same SP, not 1 LUN to SPA and another to SPB.

But you'll end up with more than 1 LUN per RG. These are ATA drives we're talking about ! More than 1 LUN per RG will give you a more random like I/O pattern and thus greatly decreased performance !! I would seriously consider staying with 1 LUN per RG, to make sure you avoid having random I/Os as much as possible.
Having said this I realize that we're aiming for 2TB LUN's and with those huge 1TB spindels there's the RAID5(2+1) option and making a meta out of 2 of such LUN's, but that would be very bad for performance indeed. RAID5(4+1) will give you the same amount of room (2TB) and only 1 parity instead of 2.

Hi Folks,

I am glad some of you are referencing the CLARiiON Best Practices Guide (BPG) - that is a helpful ref:

http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/H2358_clariion_best_prac_fibre_chnl_wp_ldv.pdf


But, I see there are some distortions creeping into the discussion.

"500 GB max LUN for ESX" - that would be an ESX-specific rule of which I am not familiar, please check the best pracs doc for VMWare.

"you have to create a single LUN on a RaidGroup " Not at all true. Actually, our testing with CDL shows increasing overall performance with up to 4 sequential streams per RAID group, on ATA. It does not matter if all 4 are in 1 LUN or in 4 separate LUNs, holding all else equal (like, how far apart they are in the filesystem).

"So assuming I want to create 2 TB meta LUNs, what is the best way to configure?" That really depends on yuor workload. I'd expect your ESX load to be random, and in this case the BPG has helpful hints for sizing the disks to the workload.

"The EMC document CLARiiON Best Practices for Fibre Channel Storage states that RAID Groups of more than ten drives results in decreased performance." Please re-read the doc's section on RAID Group Sizes, there are some comments specific to high bandwidth. Basically, for bandwidth you're best concentrating to fewer drives. We get more sequentiality from the disks that way

"BUT: meta luns perform better, ... more than 1 RG and... more physical spindels" Metas can perform better, but it depends on the workload, and how many other workloads you end up sharing those drives with. Please review the section on MetaLUNs in the BPG

"A single R5(14+1) raid group with 2 TB LUNs carved out of would would provide much worse performance than 3 x R5(4+1) raid groups with 2 TB meta LUNs carved out, right?" No,not correct. For random it would do just fine. Rebuild would be long, though. Covered in the BPG...

Also, regarding comments about assigning LUNs from ATAs to SPA vs SPB, there are some guidelines for the ATA-DAE implementation (Special DAE for ATA drives only found in CX-series arrays). Covered in...the BPG!

Dave,
Thanks for jumping in here and setting us straight