Re: Informix on HP AutoRaid

In article <6ppbuk$sog$1@news.xmission.com>, Marek Miloszewski <marek.miloszewski@zes.szczecin.pl> writes > >We are about to by a new HP9000 with Informix 7. In plans it has HP AutoRaid >disk array, but our application producer says Informix has its own mirroring >tools which we cannot turn off (?) and its better to use set of single >disks. On the other hand AutoRaid (in HP papers) looks like quite good less >administration disk array. Unfortunately AutoRaid is very new product (not >installed in Poland jet) and nobody can explain concrete how it works. >Have any of you any experiences with HP AutoRaid? Should we try it or rather >chose single disks or HP model 20 disk array? And one more: is it true that >Informix 'prefers' more small disks instead of less big? > >Thanks for any help >Marek > > I have been using two HP AutoRaid disk arrays since March of this year, prior to that we were using Model 20 and some normal single disks. I can honestly say that I've been very impressed overall with the AutoRaids - the beauty of it all is that you don't have to know too much about them, they're almost "fit and forget" devices. However, no techie worth his salt is going to leave it at that! I recommend reading the two manauals for background info and some tips. These manuals come shipped with the AutoRAID itself, I was unable to track them down beforehand - which would have been useful for investigative work. We use AutoRAID Model 12H, so I can only give you the part numbers for that type: HP Disk Array with AutoRAID: Model 12H User's Manual (C5445-90901) HP Disk Array with AutoRAID: Model 12H System Administrator's Guide (C5445-90902) Basically the AutoRAID will try to service all I/O requests from the cached controllers (these being much more intelligent and better than the Model 20 ones). It will then try to store the data in a RAID 1/0 area (especially the write intensive blocks), with least infrequently used and/or read intensive blocks on a RAID 5 area. RAID 1/0 performs a lot better than RAID 5. The main points I can make for the AutoRAID are: 1. If you can justify it, go for the Model 12H (allows use of fast 9Gb disks) with full complement of cache on two controllers; 2. Only create DISK LUNS (which appear to the OS as disk devices) which will allocate 50-60% of the available space. This will more or less guarantee that most of your data will be held as RAID 1/0. By going for the 9Gb disks (or larger if necessary) should allow this extravagance (PS disks are relatively cheap nowadays); 3. Up to 8 LUNS can be created, so use 5 or 6 (as opposed to just using 1 or 2 very large LUNS), and will leave a couple spare for future LUNS/emergencies. This will give 5 or 6 devices visible to the OS (LVM) and allow flexibility in placing devices across the two controllers (for added performance). 4. Again if you can justify it, try to fully populate all the slots. This will allow AutoRAID to internally spread the data blocks over more disk spindles (max 12 for Model 12H). Improves performance, increases the number of potential I/Os that can be performed at the same time. 5. Always use a hot-spare. I've had two disk failures occur at different times, and by the time I'd noticed it had already rebuilt the failed disk onto the hot-spare. Without that, performance would have been hit as it limped along until the failed disk had been replaced. 6. Use HSC SCSI adapters if possible, they outperform the HP-PB ones greatly (especially on the newer K range. Connect the two AutoRAID controllers to their own HSC SCSI adapter card, allowing LVM to dual- path the disk devices for greater resilience. With the older Model 20 "Nike" arrays, we just configured most of them have into 4 or 5 RAID 1/0 groups (on either side). Once set up we rarely changed them, whereas the HP AutoRAID will dynamically tune itself - disk blocks that have mostly been read will be kept in RAID-5 for example (leaving the RAID 1/0 areas in the AutoRAID for the most active write intensive disk blocks). On our Development machines, where performance is less of an issue, you could allocate much more of the capacity of the AutoRAID (even up to 80-90%). A big benefit with AutoRAID over single disks is you just don't have to worry about a disk failure. It's all handled at the hardware level. With standalone single (normal) disks, one has the admin overhead of trying to place disks on separate SCSI cards, more planning is required and you have more work to do should a disk fail. Informix probably could perform really well using small disks on a well-thought out system, but what about the overhead of Informix mirroring (I'm afraid I have no experience of using Informix mirroring), reads may perform quicker but will write performance be a drag? Does your application do many writes? The cost of all those single disks, plus extra SCSI cards, plus OS mirroring software costs may push you to hardware arrays. I think that's about it, I hope I haven't missed anything crucial... TTFN -- Simon Barber