Adding drives to a VA7400: Performance Implications

[Raghava Reddy]

We have an VA7400 with 35 18GB drives (19 in RG1 and 16 in RG2) configured in RAID1+0 mode. This storage has been fully allocated to the two existing LUNs.

I now plan to add 10 more 73GB drives (5 in each RG), and create two new LUNs. My question:

Bearing in mind that this is RAID1+0, will the two new LUNs each be striped across just four 73GB drives (one 73GB drive in each RG being reserved as a hot spare) ? This appears likely if the LUN-to-drive mapping is static and determined at LUN creation time. This implies performance for the new LUNs will be inferior to the existing LUNs.

Looking for corroboration or refutation ... thanks.

Raghav

[Ted Buis]

The VA is a virtual array, so the load will balance across the new disks. You new LUN is not limited to the new disks. If you are using HP-UX and only have the two luns, have you done anything about setting the queue depth higher than the default of 8? Also, you might what to have different mount options on different LUNs so more LUNs can be useful in some cases.

[Raghava Reddy]

Ted,

Thanks for the response. Your reply indicates that existing LUNs are re-mapped when when new disks are added, which is entirely possible. However, I'd feel a little more comfortable if you can point me to HP documentation that confirms this.

And yes, I use a queue depth of 32 for the LUNs on this array. Most of our other VAs are configured with multiple LUNs ... this particular array is an exception.

[Raghava Reddy]

Had some input from a HP engineer which is in accord with my original hypothesis: Since the existing disks are fully allocated, the new LUNs will only be striped across the new disks (existing LUNs will NOT be re-mapped).

In the absence of HP-authorized documentation, I'll have to go with the HP-employed engineer.

[Uwe Zessin]

I have found an old white paper from 2001: 5980-8568

It says:
""The HP virtual array accepts new disks while the array is up and running and accepting I/Os as with some higher-end traditional arrays. However, the HP virtual array takes it one step further. Once the disk is inserted, the array automatically includes that disk into the existing disk space and stripes all LUNs across that disk. This means that even without the creation of any additional LUNs, the array performance will improve because of the additional available spindle. Only the HP virtual array automatically adds the new disks to existing LUNs. Further, any newly created LUNs are also automatically spread across all the disks in the array, including the additional disk.""

[Ted Buis]

From the same white paper (attached), but maybe this only applies to AutoRAID, and not forced RAID 1+0. "the virtual array automatically performance tunes the array 24 hours per day, 365 days per year. The RAID level is matched with the workload. Data with access patterns that would benefit from RAID 1+0 storage are automatically directed to a RAID 1+0 section of the array. Infrequently used data, or data whose access patterns match RAID 5DP performance characteristics, are directed to the RAID 5DP section of the array.
These are the same rules that a storage expert would use to optimize an applicationâ s
performance. In addition, like an expert storage administrator, any changes to the
configuration are made only during low-usage periods or when disks are added. Note:
when disks are added, the existing LUNs are automatically extended across the new disks.
This normally requires data movement within the array. However, administrators have the
option to postpone this data movement by simply turning off the arrayâ s â Auto-Includeâ

[Ted Buis]

Forgot to attach the file.

[Ted Buis]

This additional whitepaper might also be useful.

[Raghava Reddy]

Thanks ... I read the paper. The bit about striping existing LUNs across the new disks probably applies only to AutoRaid configurations ... Ted seems to feel the same. If I find the time, I'll run some I-O benchmarks once we add the new disks (won't be for another month). Comparing results for the old and new LUNs should shed some light on this issue.