Informix Mirroring
I have a question about mirroring. I am running Informix 7.24 on a 3 processor Sun ES3500. Disks are fiber attached drives. My question is: (1) Which is better --- Informix mirroring or mirroring thru Veritas volume manager? (2) If I use Informix mirroring are there any problems? Someone told me that if the Primary Chunk is lost, then the database will crash. The database will only continue running if the Mirror Chunk is lost. Is this true? ___________________________________________________________ Jay Aymond Community Coffee jaymond@communitycoffee.com
Jay Aymond wrote: > I have a question about mirroring. > > I am running Informix 7.24 on a 3 processor Sun ES3500. > > Disks are fiber attached drives. > > My question is: > > (1) Which is better --- Informix mirroring or mirroring thru Veritas volume > manager? > I personally prefer (in this order): HW-mirroring (maybe in RAID system), OS-mirroring, Informix mirroring > > (2) If I use Informix mirroring are there any problems? Someone told me that > if the Primary Chunk is lost, then the database will crash. The database will > only continue running if the Mirror Chunk is lost. Is this true? > If this was true - what would be the sense of Informix mirroring? > > ___________________________________________________________ > Jay Aymond > Community Coffee > jaymond@communitycoffee.com -- Helmut Leininger Bull AG / Vienna Open Systems Support Email: h.leininger@bull.at helmut.leininger@bull.net This opinion is mine and not necessarily that of my employer. No guarantees whatsoever.
Helmut Leininger wrote: > > Jay Aymond wrote: > > > I have a question about mirroring. > > > > I am running Informix 7.24 on a 3 processor Sun ES3500. > > > > Disks are fiber attached drives. > > > > My question is: > > > > (1) Which is better --- Informix mirroring or mirroring thru Veritas volume > > manager? > > > I personally prefer (in this order): HW-mirroring (maybe in RAID system), > OS-mirroring, Informix mirroring Informix's own testing, though several years old, agrees with you. Just stick non-OS software mirroring between OS-mirror and Informix-mirror. Veritas is more flexible than Informix mirroring and slightly faster to boot. Also Informix will alternate reads between the two mirrors but it ALWAYS writes to the primary first. Some hardware and software mirrors write to both simultaneously which can give better performance. Informix ties up an IO thread for two operations for it's own mirrors while an external mirror requires only one Informix operation so the engine performs better even if the mirror's performance is identical. > > (2) If I use Informix mirroring are there any problems? Someone told me that > > if the Primary Chunk is lost, then the database will crash. The database will > > only continue running if the Mirror Chunk is lost. Is this true? > > > > If this was true - what would be the sense of Informix mirroring? I've never heard of or seen the problem you describe. If the primary chunk is down the mirror is used if one is available. Art S. Kagel
This suggests (for me) some questions. I'm sure these are naive. >Art S. Kagel <kagel@bloomberg.net> wrote in message news:38172AA1.2466A96@bloomberg.net... > Helmut Leininger wrote: ... > Informix's own testing, though several years old, agrees with you. Just > stick non-OS software mirroring between OS-mirror and Informix-mirror. > Veritas is more flexible than Informix mirroring and slightly faster to > boot... Would Disksuite be an example of"OS-mirroring" or "non-OS"? Are you suggesting 3 layers of mirroring? i.e., mirroring at OS, some "intermediate" form of mirroring, and Informix mirroring? I don't understand what 3 layers of mirroring would mean or do. Would it be a complex metadevice? In our warehouse application, there is very little writing. If I'm understanding you, Art, you may be suggesting that DiskSuite mirroring might be preferred over Informix mirroring, but Veritas might be preferred over DiskSuite. I don't want to put words in your mouth, this is just a perception check. > ... Informix > ties up an IO thread for two operations for it's own mirrors while an > external mirror requires only one Informix operation so the engine > performs better even if the mirror's performance is identical. Would there be any net gain? If work is just shifted from the Informix engine to OS or other software mirroring technique, does it affect the view "from outside"? In other words, would it be like transferring coins from my right pocket to my left, and taking comfort that my right pocket will last longer, but forgetting about effect on left pocket? Of course, there may be a situation in which the right pocket is more important than left, so the analog limps a little. Sign me a little confused, David Grove
David Grove wrote: > > This suggests (for me) some questions. I'm sure these are naive. You mean I have to THINK about what I said? Good thing it's not too early in the day. > >Art S. Kagel <kagel@bloomberg.net> wrote in message > news:38172AA1.2466A96@bloomberg.net... > > Helmut Leininger wrote: ... > > > Informix's own testing, though several years old, agrees with you. Just > > stick non-OS software mirroring between OS-mirror and Informix-mirror. > > Veritas is more flexible than Informix mirroring and slightly faster to > > boot... > > Would Disksuite be an example of"OS-mirroring" or "non-OS"? Disksuite would be software mirroring. Linux kernel mirroring would be OS mirror, DEC Ultrix also had mirroring and striping implemented in the kernel. Kernel based or OS mirroring is often incrementally faster than software mirroring because it may use one less layer of device drivers and run in system space rather than user space. > Are you suggesting 3 layers of mirroring? i.e., mirroring at OS, some > "intermediate" form of mirroring, and Informix mirroring? I don't > understand what 3 layers of mirroring would mean or do. Would it be a > complex metadevice? In our warehouse application, there is very little > writing. Then the following comments about write performance would not apply to you. > If I'm understanding you, Art, you may be suggesting that DiskSuite > mirroring might be preferred over Informix mirroring, but Veritas might be > preferred over DiskSuite. I don't want to put words in your mouth, this is > just a perception check. Ummm. I'd class Disksuite and Veritas at the same level, though I suspect that Veritas is a better product. Both are addons not OS level drivers. What I am saying is that on, say Linux, with kernel OS mirroring I'd expect that the OS mirror would outperform Veritas, if and when it becomes available for Linux (maybe it already is?). > > ... Informix > > ties up an IO thread for two operations for it's own mirrors while an > > external mirror requires only one Informix operation so the engine > > performs better even if the mirror's performance is identical. > > Would there be any net gain? If work is just shifted from the Informix > engine to OS or other software mirroring technique, does it affect the view > "from outside"? In other words, would it be like transferring coins from my > right pocket to my left, and taking comfort that my right pocket will last > longer, but forgetting about effect on left pocket? Of course, there may be > a situation in which the right pocket is more important than left, so the > analog limps a little. OK, I thought I'd lose someone with this one. Yes it IS like transferring coins from the right pocket to the left, I have just as much money as before, no more, no less. However, I can now have my right hand go back to searching the right pocket for more quarters while my left hand pays the bill. I get done faster! If the AIO VP or KAIO thread is waiting for two disk writes to complete before picking up another I/O job, and all of the IO threads are doing the same thing, then less work gets done in the engine than if a single IO is executed and the two writes are simultaneous. Remember if you have structured your mirrors for maximum redundancy and uptime then each member of a mirrored pair is on a different drive and controller so that the two write CAN indeed happen in parallel, simultaneously, there is no reason to wait for one to complete before launching the other and this INCREASES the safety of the operation since you reduce the possibility that the machine will crash before the mirror drive can be written requiring a mirror rebuild on reboot. By pushing the operation to the OS or hardware the IDS threads can be available more quickly to perform other reads and writes perhaps from other drives. Art S. Kagel