What would you consider a RAID-10 volume that is "too large"?

[David Sullivan_7]

I'm about to implement my first entry-level SAN for a customer that has outgrown their SmartArray 6404 and MSA30 with 14 drives in a RAID-10 array. GlancePlus has shown that they are severely disk-bound in their application (the array runs 100% busy almost all day long), so we have proposed migrating them to an EVA with 28 drives. Capacity is not an issue for this customer: our application is very I/O heavy and they have deep IOWAIT queues. The move to the EVA is simply to increase the spindle count to keep up with the I/O load. In studying some of the postings on the forum where you good folks have written about your significantly larger storage subsystems than I'll be dealing with, some have noted that they create multiple RAID LUNs and then create OS volumes on top of these LUNs. So my question to this august body would be: what determines that point? Would a 28-drive RAID-10 array be too large for an EVA? Should I split it into two (or more)arrays and combine the LUNs into a single filesystem at the HPUX level?

TIA!

[Rob Leadbeater]

Hi David,

If you're moving to an EVA, then the concepts of RAID-10 that you're familiar with on the SmartArray, aren't particularly relevant...

For a small system like you're proposing, then data would be automatically distributed across all the spindles on the EVA, so you should see a performance boost there, plus the increased cache will help.

How is the storage currently laid out on the SmartArray controller ?

Cheers,

Rob

[Víctor Cespón]

Hi, please note that the new EVAx400 supports SSD drives. If your application is I/O limited and needs fast speed and not much space, you may ask prices for 8 SSD and compare that to 28 15K mechanical drives.

Each SSD can do thounsands of operations per second, while a mechanical disk maxes out at 160.

Of course you can also have FC drives, or even FATA drives on the same EVA to store less frequently accessed data.

[Patrick Terlisten]

Hello,

the EVA doesn't handle their VRAIDs like a traditional RAID controller (for example the Smart Array Controllers). A Vdisks is distributed over all disks in a disk group. If you have a 1 GB VRAID 1 Vdisks it's distributed over all disks in a disk group, even if the diskgroup has 240 drives. In this case every disk would hold 1/240 of the 1 GB Vdisk. More disks = more performance. The EVA will boost your application performance noticeable.

As vcespon already wrote: Maybe SSDs are an interesting option for you. But you need at least an EVA4400 and you can only put up to 8 SSDs into one EVA. It's only an option if you need less then 460 GB for you application. This is because you can plug only up to 8 72 GB SSDs into one EVA and you only can use VRAID 5. If you need more diskspace for your application, you have to use normal FC drives. You should use 15k drives and use more small drives instead of fewer bigger dissk.

Best regards,
Patrick

[David Sullivan_7]

Thank you all for your input. I must say that I obviously need more education in EVA technology before I cast all of these drive arrangements into stone.

Rob, to answer your question, the MSA30/SmartArray setup was a single RAID10 array composed of 14 @ 73GB 15K RPM drives split across two SCSI channels. At the HPUX level, this was a single large filesystem. The customers database is only about 100GB, but their I/O rate is insane.

The suggestion to look into SSD's is a very good one. I'm a little skittish, however, after reading some information about the load-levelling algorithms in Intel's high-performance SSD's that would actually cause them to become *slower* over time, to the point that they eventually got slower than regular laptop harddrives.

We actually are going to be using Fibre drives in the EVA model the customer purchased to maximize performance.

So, I guess my next question would be "how should I set the drives up to maximize performance"? Being rather ignorant of how to configure EVA's, it seems the language that I'm used to communicating with (i.e. hard arrays on physical controllers) doesn't make as much sense in the SAN space.

[Víctor Cespón]

The SSD used by HP (and almost all other storage systems vendors) are enterprise class from STEC, not the ones from Intel. http://www.stec-inc.com/press/articles/STEC_goes_all_the_way_with_EVA.pdf.

The ones from Intel are not bad, but internally are different, they are MLC with a 64 MB RAM cache, not SLC. SLC drives can sustain 10 times more write cycles than MLC and speed does not degrade with age. Also, all SSD have a 5% or so of extra space to replace sectors when they no longer work as expected.

Also, consider that a SSD is less likely to fail, as it has no moving parts and does not get very hot like a 15K drive. They even can warn when their lifespan is about to end, and be proactively replaced. I still don't know what will be HP's policy regarding this, no SSD has reached that point yet.

[Simon Setina]

Hi David,

As you need only 100GB of space, than you can think about Texsas Memory Systems (www.suuperssd.com) RamSan400, whic can be equpied up to 128GB of Dynamic RAM, not Flash. RS400 can reach up to 600K IOPs.

Simon