New System Config and DS8800

Hello Throwing out feelers on some questions regarding the setup of Informix OLTP systems on the new DS8800 Storage. We had drawn up a RAID 10 design for both data and logs and received feedback that RAID 5 would work out just fine. Through the years, the literature, conferences and advice from IBM has been that RAID 10 is best for logs and the only drawback for 10 for data is cost (and that in general, warehouse is more conducive to RAID 5 and OLTP is RAID 10). Our current system is RAID 5 but we were thinking going to RAID 10 is a step up. Maybe there are some things with newer technology with cache, etc, etc that we are unaware of that changes the equation? Also, we are wondering if any others have already looked at having several OLTP applications running on one large extent pool of disks, spindles, etc versus compartmentalizing the different OLTP systems into smaller sets of disks, spindles, etc. Obviously bringing in hardware and running benchmarks is the emprical proof in the pudding but this is more at the stage of drawing up plans and architecture. The main cash cow is on AIX 5.3 and it is an HDR pair running on a HACMP cluster. Thanks, Tom

NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! NO RAID5!!!!! Got that? Read my RAID 5 Rant in the June edition of my Blog ( http://informix-myview.blogspot.com/2010_06_01_archive.html) if you want the details about why NO RAID5, or better yet, go to the BAARF web site ( www.baarf.com) and read about the evils of RAID5 from myself and several experts including two well respected Oracle engineers. Also while you are there, go to the membership pages and read the blurbs from dozens of people who lost data and/or time from RAID5 failures. NO RAID5!!!!! Performance asside, RAID5 is not safe at any price! Not worth the really minimal cost savings given the cost of storage spindles today. Go RAID10 all the way! Look RAID5 was envisioned at a time when disk storage was prohibitively expensive. A 5MB disk pack cost many thousands of 1980's dollars and RAID1 mirroring was just too expensive then so RAID2, 3, 4, & 5 were devised. Today I can buy a 3TB drive for under $100US, yes, the high speed drives used in an EMC Symetrics or IBM DS8000 array are significantly more expensive than that. Are they more expensive than the time it will take to recover your system when the RAID5 fails catastrophically? Is it more expensive than the lost business? Lost customer confidence? Chase online banking was offline for more than three full days a few months ago when their storage systems failed and had to be restored from archives and transactions replayed. Just the cost of manning the complaint lines was probably more than the difference between the cost of RAID5 & RAID10. Was it RAID5 that caused the crash? I don't know, but I seriously doubt that it was RAID10 since the probability of a catastrophic failure of a RAID10 array resulting in data loss is miniscule. On using a large disk array as a pool of space for multiple systems, I am generally against it. IF you need maximum performance and maximum control over each server's resources to insure the performance you need, then each server should have its own dedicated storage. Going further each server should really have multiple independent storage arrays to separate rootdb from physical logs from logical logs from highly active data dbspaces from index dbspaces from low activity data dbspaces from historical data dbspaces, etc. I know that this is not practical in the real world today, but that's the ideal. That all said, you may not require maximum performance. If your peak load for the busiest dbspaces on your servers is each pulling a couple of hundred IOs per hour then several servers can happily share storage with little or no interference. You have to know your systems and, as you point out, you have to test everything. (OK everything except RAID5. No need to test that. It's a boondoggle, stay away from RAID5 for any data that you care about.) Art Art S. Kagel Advanced DataTools (www.advancedatatools.com) IIUG Board of Directors (art@iiug.org) Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on my employer, Advanced DataTools, the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Thu, Dec 23, 2010 at 3:24 PM, Tom Lehr <tomcaml@gmail.com> wrote: > Hello > > Throwing out feelers on some questions regarding the setup of Informix > OLTP systems on the new DS8800 Storage. We had drawn up a RAID 10 > design for both data and logs and received feedback that RAID 5 would > work out just fine. Through the years, the literature, conferences and > advice from IBM has been that RAID 10 is best for logs and the only > drawback for 10 for data is cost (and that in general, warehouse is > more conducive to RAID 5 and OLTP is RAID 10). Our current system is > RAID 5 but we were thinking going to RAID 10 is a step up. Maybe there > are some things with newer technology with cache, etc, etc that we are > unaware of that changes the equation? Also, we are wondering if any > others have already looked at having several OLTP applications running > on one large extent pool of disks, spindles, etc versus > compartmentalizing the different OLTP systems into smaller sets of > disks, spindles, etc. Obviously bringing in hardware and running > benchmarks is the emprical proof in the pudding but this is more at > the stage of drawing up plans and architecture. The main cash cow is > on AIX 5.3 and it is an HDR pair running on a HACMP cluster. > > Thanks, > Tom > _______________________________________________ > Informix-list mailing list > Informix-list@iiug.org > http://www.iiug.org/mailman/listinfo/informix-list >