The AMS200 we had to buy a rack for, which we hadn’t planned on. The assumption was made that a disk subsystem would come with a rack, based on past experiences, but HDS tried to save us some cost by not making us buy their rack at like 3x the cost for a standard server rack. It fits in a standard 19 in. server rack with your everyday PDUs. You just need two to four 20 amp circuits (for our PDUs) and you’ll be good to go. You need a couple ethernet connections and only two fibre channel connections. HDS will tell you that it has four fibre channel connections, but once you connect it to a fabric, then you are only able to use two ports. They actually have a fibre channel hub built into the processors so that you are only going to get full bandwidth for a single 2Gb port per processor. Technically you can leave the port in FC-AL mode and connect all four ports to the fabric, but you’re only going to get bandwidth for a single port per processor. Once you tell the ports that they are in fabric mode, one of the two ports per processor gets disabled. I griped about this quite a bit and have been told that all the new 4Gb processors don’t have the fibre channel hub any longer and you can run both ports per processor at full bandwidth. Well this would have been nice to know prior to the order going out. Oh well.
So once we got the AMS up and running it was a piece of cake getting setup. We have configured the 9 fibre channel drives in a 7+1 RAID 5 set with one hot spare. We have four shelves of 250GB SATA drives that I configured as four 12+2 RAID 6 sets with one hot spare per shelf. I know, you’re thinking to yourself this is over kill on the spares, but using the 12+2s what else was I supposed to do with them. Later down the road when we order more disk, I’ll probably make two of the four part of another RAID group and we’ll get them back. Evidently there is some debate at HDS about the “one spare per shelf” rule. From what I understand the track record for the drives is so good that the field engineers aren’t replacing near as many as the marketing guys say you should expect to replace. The SATA drives have been around for a few years now and I’m sure that they have worked out the major bugs by now. We’ll see. They’ve got to be better than the drives we’ve been getting from EMC in the Clariions.
So back to the 12+2 performance. We did some performance testing with some DB2 database backups going to two 6+1 RAID 5 RAID groups and two 12+2 RAID 6 RAID groups. We saw little, if any, difference in the performance between the RAID group configurations. Upon further questioning to HDS and their backline engineers, the ASICs are now computing the parity calculations and they can handle way more than what we’re pushing or could push into those RAID groups. I could have gone larger with the RAID 6 groups, but I decided to keep them contained to each shelf rather than spanning shelves. Just a personal preference for now, no technical reason behind it.
The only other gripe I have about the AMS200 is the lack of being able to copy Host Groups or LUNs from one port to another. This seems like it would be standard, since they have had it for the 9980V for at least 3+ years. It’s not a huge deal, but I don’t think they did many end user studies for the GUI for this subsystem.
The AMS200 has been performing admirably so far and we are moving data from the cx600 to it daily.