VA7100 controller settings.

[Don Watson]

I need to configure a VA7100 array. I'm using cvui, and have set the array for RAID1+0, however there are several configuration parameter options that I don't understand. The help doc does not explain the trade-offs of choosing one opton over another, and some of the defaults don't make sense. For example:

1. Hot spare - Using RAID1+0 I have in effect 1 hot spare for each primary disk already. Why would I want to reserve a disks worth of space for yet another hot spare? Is this just to cover the time between a disk failing and getting it replaced, so that mirroring is never interrupted? Is there any other downside to setting this to none?

2. Optimize Policy - It sounds like Optimize Policy is only applicable in an AutoRaid setup. Does this setting have any relevance at all for RAID1+0?

3. Pre-fetch - Why would the default be set to off? Why would anyone want to disable read ahead? Is there a performance penalty, or some other reason why you might not want this enabled?

[Vincent Fleming]

Hi;

To answer your questions...

1. Yes, it is to cover you from when a drive fails until you replace the drive. It's an added safety feature.

2. Optimize Policy is really only valid in AutoRAID mode.

3. There are a number of parameters that actually do nothing (such as turning write caching off!), but are there for compatibility reasons. I can't find my list of things, but I think Pre-fetch is one of those.

Good luck!

[Don Watson]

Ok, I'm with you on Hot Spare and Optimize Policy. I also read in the doc where some of the parameters aren't really functional, and are there for compatibility with O/S's that require certain settings, however, Pre-Fetch is not documented that way. Not saying it really is functional, but it's not listed as being one of the useless compatiblity options.

[Vincent Fleming]

I've been reading up for you...

Apparently the pre-fetch does indeed do something; disabling read-ahead can increase performance in highly random, small-block read I/O situations. Mostly the option to turn it off is used for benchmarking, so you normally won't want to turn it off as in the real-world, I/O patterns like that would be very rare.

Good luck!

[Vincent Farrugia]

Hello,

The scope of having a hotspare is simple. It's true; in RAID 1+0, you have mirroring, and effectively another disk for all disks. What hotspare does is... whenever the machine senses a disk failure, it rebuilds all the data to the hotspare immediately. If you had no hotspare, the system would still function and no data would be lost, however, if you lose another disk, data in that LUN would be lost. Hotspare enables the machine to rebuild the LUN to the hotspare disk if a failure occurs, thus bringing the machine in a non-urgent state again.

HTH,
Vince

[Don Watson]

Thanks for the replies. Very helpful.

One last question, if I turn on "Hot Spare", allocate my LUN's, and then at some later time decide I need to use that hot spare disk space, can I just turn off the hot spare option and have it return the space for LUN allocation? If so, would this happen on the fly or require any downtime?

I'm guessing I can do this, but would really like to know for sure up front.

[Vincent Fleming]

Yes, you can change your spare setting at any time.

Changing to no spares will behave as you expect - the space will be able to be allocated as a LUN without any downtime.

Good luck!

[Don Watson]

Ok. But I guess I lied about the "one more question" thing.

What exactly is "automatic"? Doesn't the concept of hot spare imply reserving an amount of space equal to the size of at least a disk?

Also, I checked my available space for allocating LUN's (with hot spare set to none), then set hot spare to "largest disk", checked the available space again and it was exactly the same as before. How is that possible?

If a good explanation of hot spare is documented somewhere, a link would be a fine answer. Thanks.

[Vincent Farrugia]

Hello,

For starters, follow this link:

http://www.hp.com/cgi-bin/cposupport/get_doc.pl?SNI=hpsurestor30307&LC=information_storage&Tfile=lpg35046

For advanced, follow this link, pg. 52:

http://www.hp.com/cposupport/manual_set/lpg28817.pdf

You can also read the rest of the manual of course :-)

HTH,
Vince

[Don Watson]

Thanks, but the first link is quite vague and leaves many questions unanswered. When I try the second link I get an error that the file can't be downloaded.

I have actually read every manual I can get ahold of, including every one that HP support has directed me to, but nothing I can find so far fully explains hot spare usage, or even gives a useful definition of how each option works for each of the different RAID levels.

[Roger_22]

Don,

The confusion is about how the VA implements hot spares. Hot spare in the VA, called Active Hot Spare, is implemented as virtual spare space. That is, there is not a physical disk(s) associated with the spare space. The spare space is distributed across all the disks in the RAID group. Visualize a group of beakers that represent the disks. Data to the array is like water to the beakers. The water (data) is naturally distributed evenly to all the beakers. If a beaker was to fail, the water in that beaker can be re-distributed to the un-use capacity of the remaining beakers. Enabling Active Hot Spare merely assures that the beakers will have the reserved space available in case of a beaker failure.

So, why don???t you see a different in capacity for RAID 1+0 by enabling and disabling Active Hot Spare? The VA has one more trick to use. If there is insufficient free space in the beakers for the failed beaker???s water, the VA knows it can temporarily convert some of the inefficient RAID 1+0 capacity to the more efficient RAID 5DP capacity ??? thus freeing up sufficient capacity for the defective beaker???s water. Once the broken beaker is replaced, the water is re-distributed back to the new beaker and the RAID 5DP space is converted back to RAID 1+0.

As long as AutoRebuild is enabled, the VA will always attempt to recover from a disk failure ??? one of its prime directives ??? don???t lose data.

The beauty of this implementation is its protection power and its flexibility. It???s all controlled with the Enable Active Hot spare switch and difference between the capacity allocated to LUNs and the physical capacity of the array ??? the free space. By reducing the reserved LUN capacity, the array can withstand multiple sequential disk failures, or, in the case of your RAID 1+0 configuration, not convert any of the data to RAID 5DP first.

[Don Watson]

Now we're getting somewhere.

Good explanation, but I'm still not clear about a couple of details. I get the part about "automatic" using a temporary RAID5DP conversion if a disk fails, but how does the "largest disk" option work, and under what circumstances would that be a better choice. What I'd love to find is a white paper describing the available scenarios with the trade-offs for each setting, but I've been unable to locate any such info.

If you are getting your facts from doc that's online I'd be perfectly satisfied with that rather than pestering everyone.

Thanks.

[Don Watson]

One more point, I've got a VA 7100 rather than a 7400 if that matters.

[David Bell_1]

Don,

Not sure if you've seen all the white papers here. However, these cover the mechanics of RAID5DP. Hopefully, this will help you along,

Dave

http://www.hp.com/products1/storage/products/disk_arrays/midrange/va7100/infolibrary/index.html

[Roger_22 · Accepted solution]

Hey, if you want any more answers you???ll need to assign some points! What, do you think I???m doing this for nothing?

The VA array allows any combination of supported disk capacities. ???Largest Disk??? just means it???s reserving the spare space for the largest disk in the array to fail ??? dah! The other missing option would be to reserve space for only the smallest disk to fail.

In RAID 1+0 mode, it really does not matter if you select any Active Hot Spare setting or not. If AutoRebuild is enabled, the array will continue to ???rebuild??? on each successive disk failure until it cannot make space (by converting data to RAID 5DP if necessary) to complete a rebuild. Technically, the last rebuild will stop where it is when it runs out of space and never complete ??? it???s just waiting for someone to add another disk.

The only real choice you have to influence the algorithm (beyond disabling AutoRebuild) is to decrease the reserved LUN allocation sufficiently so that on the first disk failure there is sufficient un-reserved capacity that the rebuild can complete without creating any tempory RAID 5DP space. This will keep everything in RAID 1+0 after the rebuild. For some workloads the RAID 5DP may cause performance issues. Remember, the fewer the number of disks in the Redundancy Group, the greater the percentage of the capacity needs to be converted to RAID 5DP to create a disk???s worth of free space. If you only have five disks in the Redundancy Group, than almost the total capacity of the RG needs to be converted to RAID 5DP to create sufficient space. If you have 20 disks in the RG (this requires a VA7400), then only a small percentage of the array capacity is converted to RAID 5DP.

The bad news is that there is no way to determine specifically (the array won???t tell) how much space to under allocate the array so that a rebuild won???t create RAID 5DP space. The good news is that it really doesn???t matter. Guessing is close enough. Cut back the LUN allocation by the capacity of the largest disk in the array. This may not be enough, but only a small amount will be in RAID 5DP if any (the explanation for that is yet another note)

I know of no white paper on this subject.

Oh, always enable the new pre-fetch algorithm. It is far more efficient at pre-fetching and with almost no downside.

[Don Watson]

Sorry, I'm new to this forum and didn't get the points thing.

That's very good info, and does help me make decisions about the config settings.

Thanks for all your help.