2-disk layout
A user asked how to lay out IDS 9.2 on two 9GB disks (rootdbs plus extra chunks on disk 2, OS/apps/logs on the boot disk) and what hardware RAID-1 would require. No one really addressed the specific 2-disk layout; the thread turned into a RAID tutorial. A poster listed RAID 0-5 characteristics, and Art Kagel corrected it: RAID1 reads are as fast or faster than single disks, RAID3/4 are fine for databases, and RAID5 should never be used for databases (slow writes, parity not verified on read, and far greater risk of permanent data loss after a partial media failure). His recommendation was RAID10 (striped mirrored pairs). The original layout question itself was left unresolved.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
I need to come up with a 2-disk (2 9GBs) layout for IDS 9.2. Let's say I have a database of up to 6GBs. I have identified 4 other chunks (each with its own slice of 50 MB) that can be added to the rootdbs. I am thinking of putting the rootdbs and all its chunks on the second disk. The boot disk is mainly used for the OS, applications, and physical/logical logdbs. Does it sound good ? Recommendations are appreciated. Thanh Sent via Deja.com http://www.deja.com/ Before you buy.
In article <8muoik$ttd$1@nnrp1.deja.com>, Thanh <tqma@my-deja.com> wrote: > I need to come up with a 2-disk (2 9GBs) layout for IDS 9.2. > Let's say I have a database of up to 6GBs. I have identified > 4 other chunks (each with its own slice of 50 MB) that can > be added to the rootdbs. I am thinking of putting the rootdbs > and all its chunks on the second disk. The boot disk is mainly > used for the OS, applications, and physical/logical logdbs. > > Does it sound good ? Recommendations are appreciated. Also, being new to hardware RAID, what do I need to have RAID-1 with the above configuration ? Thanks, Thanh Sent via Deja.com http://www.deja.com/ Before you buy.
RAID0 - Striping without Parity Striped Disk Array without Fault Tolerance - minimum 2 disks - very, very fast. Sucks if you don't have a good backup all the time. RAID1 - Mirroring and Duplexing - minimum 2 disks (duplexing needs 2 controllers). High tolerance for faults, slow but oh, so, reliable. RAID2 - Hamming Code ECC - great when your clients are reading and writing large data blocks at high data transfer rates, but smaller block reads are inefficient. Read, modify, and write operations required for small block write operation also result in poor performance. RAID3 - Striping with Parity - minimum 3 disks. Very fast, and good at fault tolerance. Excellent array for video and imaging or any application requiring high througput. Complex to maintain. RAID4 - Independent Data disks with shared Parity disk - requires 3 disks. Slow writes, fast reads! Complex to setup and maintain. RAID5 - Independent Data disks with distributed parity blocks - minimum 3 disks. Highest Read speed, medium write speed. Recommened for: File and Application servers Database servers WWW, E-mail, and News servers Intranet servers Most versatile RAID level and most complex controller design and as well as difficult to design. Finally, someone correct me if I'm wrong, I don't know about the operating systems, but WinNT only supports 0, 1, and 5 _____________________________________________ "The most damaging phrase in the language is: `It's always been done that way.'" -Amazing Admiral Grace Hopper ----------------------------------------------------------- Got questions? Get answers over the phone at Keen.com. Up to 100 minutes free! http://www.keen.com
>Finally, someone correct me if I'm wrong, I don't know about the >operating systems, but WinNT only supports 0, 1, and 5 Ooops, I meant, I don't know about the *other* operating systems,.... ----------------------------------------------------------- Got questions? Get answers over the phone at Keen.com. Up to 100 minutes free! http://www.keen.com
NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 wyn441 wrote: > > RAID0 - Striping without Parity > Striped Disk Array without Fault Tolerance - minimum 2 disks - > very, very fast. Sucks if you don't have a good backup all the > time. > RAID1 - Mirroring and Duplexing - minimum 2 disks (duplexing > needs 2 controllers). High tolerance for faults, slow but oh, > so, reliable. Slow compared to what? Compared to singleton disks RAID1 is actually slightly to 50% faster on reads due to reduced head contention because most RAID1 systems can read from both drives in parallel, and about equal to singleton drives for writes. > RAID2 - Hamming Code ECC - great when your clients are reading > and writing large data blocks at high data transfer rates, but > smaller block reads are inefficient. Read, modify, and write > operations required for small block write operation also result > in poor performance. > RAID3 - Striping with Parity - minimum 3 disks. Very fast, and > good at fault tolerance. Excellent array for video and imaging > or any application requiring high througput. Complex to maintain. Good for databases too, actually. Excellent protection from partial media failure, parity read and verified on read! New hardware implementations are fast at parity without CPU resource usage. > RAID4 - Independent Data disks with shared Parity disk - > requires 3 disks. Slow writes, fast reads! Complex to setup and > maintain. Almost as good as RAID3 for databases. Excellent protection from partial media failure, parity read and verified on read! New hardware implementations are fast at parity without CPU resource usage. > RAID5 - Independent Data disks with distributed parity blocks - > minimum 3 disks. Highest Read speed, medium write speed. > Recommened for: > File and Application servers > Database servers AAAAAAAAAAAAAAAAAAAAAAAAAA! NEVER NEVER NEVER use RAID5 for databases. SSSSLLLOOOWWW writes, 50% of RAID0 write speed. Prone to catastrophic data loss due to partial media failure because parity is NEVER checked on read! > WWW, E-mail, and News servers > Intranet servers > Most versatile RAID level and most complex controller design and > as well as difficult to design. > > Finally, someone correct me if I'm wrong, I don't know about the > operating systems, but WinNT only supports 0, 1, and 5 Several controller manufacturers, ICP Vortex for one, supply controllers with NT drivers to implement RAID levels 0, 1, 3, 4, 5 and some of the hybrid RAID combos such as RAID6, RAID01 and RAID10 (my personal favorite). RAID10, stripped sets (RAID0) made of mirrored pairs (RAID1), provides the besta combination of protection, as good as RAID1, and performance, as good or slightly better than RAID0, for both reads and writes. It is almost immune to partial media failure cross-corruption and recovery time is minimal with the smallest impact on concurrent performance during recovery. RAID10 is different from RAID01 (mirrored pairs made from stripe sets) in mainly in the speed and impact of recovery and the risk of catastrophic loss during recovery due to a second drive failure. Art S. Kagel
Way to go, Art. I have tested and experienced everything you wrote. Evidently you know what are you talking about...Thanks. In article <39942B62.B7AAB426@bloomberg.net>, kagel@bloomberg.net wrote: > NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 > > wyn441 wrote: > > > > RAID0 - Striping without Parity > > Striped Disk Array without Fault Tolerance - minimum 2 disks - > > very, very fast. Sucks if you don't have a good backup all the > > time. > > RAID1 - Mirroring and Duplexing - minimum 2 disks (duplexing > > needs 2 controllers). High tolerance for faults, slow but oh, > > so, reliable. > > Slow compared to what? Compared to singleton disks RAID1 is actually > slightly to 50% faster on reads due to reduced head contention because > most RAID1 systems can read from both drives in parallel, and about > equal to singleton drives for writes. > > > RAID2 - Hamming Code ECC - great when your clients are reading > > and writing large data blocks at high data transfer rates, but > > smaller block reads are inefficient. Read, modify, and write > > operations required for small block write operation also result > > in poor performance. > > RAID3 - Striping with Parity - minimum 3 disks. Very fast, and > > good at fault tolerance. Excellent array for video and imaging > > or any application requiring high througput. Complex to maintain. > > Good for databases too, actually. Excellent protection from partial > media failure, parity read and verified on read! New hardware > implementations are fast at parity without CPU resource usage. > > > RAID4 - Independent Data disks with shared Parity disk - > > requires 3 disks. Slow writes, fast reads! Complex to setup and > > maintain. > > Almost as good as RAID3 for databases. Excellent protection from > partial media failure, parity read and verified on read! New hardware > implementations are fast at parity without CPU resource usage. > > > RAID5 - Independent Data disks with distributed parity blocks - > > minimum 3 disks. Highest Read speed, medium write speed. > > Recommened for: > > File and Application servers > > Database servers > > AAAAAAAAAAAAAAAAAAAAAAAAAA! > > NEVER NEVER NEVER use RAID5 for databases. SSSSLLLOOOWWW writes, 50% > of RAID0 write speed. Prone to catastrophic data loss due to partial > media failure because parity is NEVER checked on read! > > > WWW, E-mail, and News servers > > Intranet servers > > Most versatile RAID level and most complex controller design and > > as well as difficult to design. > > > > Finally, someone correct me if I'm wrong, I don't know about the > > operating systems, but WinNT only supports 0, 1, and 5 > > Several controller manufacturers, ICP Vortex for one, supply > controllers with NT drivers to implement RAID levels 0, 1, 3, 4, 5 and > some of the hybrid RAID combos such as RAID6, RAID01 and RAID10 (my > personal favorite). > > RAID10, stripped sets (RAID0) made of mirrored pairs (RAID1), provides > the besta combination of protection, as good as RAID1, and performance, > as good or slightly better than RAID0, for both reads and writes. It > is almost immune to partial media failure cross-corruption and recovery > time is minimal with the smallest impact on concurrent performance > during recovery. RAID10 is different from RAID01 (mirrored pairs made > from stripe sets) in mainly in the speed and impact of recovery and the > risk of catastrophic loss during recovery due to a second drive > failure. > > Art S. Kagel > Sent via Deja.com http://www.deja.com/ Before you buy.
patricia_br@my-deja.com wrote: > > Way to go, Art. I have tested and experienced everything you wrote. > Evidently you know what are you talking about...Thanks. > > In article <39942B62.B7AAB426@bloomberg.net>, > kagel@bloomberg.net wrote: > > NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 Just the result of painful experience and many long nights recovering servers from RAID5 failures. Art S. Kagel > > wyn441 wrote: > > > > > > RAID0 - Striping without Parity > > > Striped Disk Array without Fault Tolerance - minimum 2 disks - > > > very, very fast. Sucks if you don't have a good backup all the > > > time. > > > RAID1 - Mirroring and Duplexing - minimum 2 disks (duplexing > > > needs 2 controllers). High tolerance for faults, slow but oh, > > > so, reliable. > > > > Slow compared to what? Compared to singleton disks RAID1 is actually > > slightly to 50% faster on reads due to reduced head contention because > > most RAID1 systems can read from both drives in parallel, and about > > equal to singleton drives for writes. > > > > > RAID2 - Hamming Code ECC - great when your clients are reading > > > and writing large data blocks at high data transfer rates, but > > > smaller block reads are inefficient. Read, modify, and write > > > operations required for small block write operation also result > > > in poor performance. > > > RAID3 - Striping with Parity - minimum 3 disks. Very fast, and > > > good at fault tolerance. Excellent array for video and imaging > > > or any application requiring high througput. Complex to maintain. > > > > Good for databases too, actually. Excellent protection from partial > > media failure, parity read and verified on read! New hardware > > implementations are fast at parity without CPU resource usage. > > > > > RAID4 - Independent Data disks with shared Parity disk - > > > requires 3 disks. Slow writes, fast reads! Complex to setup and > > > maintain. > > > > Almost as good as RAID3 for databases. Excellent protection from > > partial media failure, parity read and verified on read! New hardware > > implementations are fast at parity without CPU resource usage. > > > > > RAID5 - Independent Data disks with distributed parity blocks - > > > minimum 3 disks. Highest Read speed, medium write speed. > > > Recommened for: > > > File and Application servers > > > Database servers > > > > AAAAAAAAAAAAAAAAAAAAAAAAAA! > > > > NEVER NEVER NEVER use RAID5 for databases. SSSSLLLOOOWWW writes, 50% > > of RAID0 write speed. Prone to catastrophic data loss due to partial > > media failure because parity is NEVER checked on read! > > > > > WWW, E-mail, and News servers > > > Intranet servers > > > Most versatile RAID level and most complex controller design and > > > as well as difficult to design. > > > > > > Finally, someone correct me if I'm wrong, I don't know about the > > > operating systems, but WinNT only supports 0, 1, and 5 > > > > Several controller manufacturers, ICP Vortex for one, supply > > controllers with NT drivers to implement RAID levels 0, 1, 3, 4, 5 and > > some of the hybrid RAID combos such as RAID6, RAID01 and RAID10 (my > > personal favorite). > > > > RAID10, stripped sets (RAID0) made of mirrored pairs (RAID1), provides > > the besta combination of protection, as good as RAID1, and > performance, > > as good or slightly better than RAID0, for both reads and writes. It > > is almost immune to partial media failure cross-corruption and > recovery > > time is minimal with the smallest impact on concurrent performance > > during recovery. RAID10 is different from RAID01 (mirrored pairs made > > from stripe sets) in mainly in the speed and impact of recovery and > the > > risk of catastrophic loss during recovery due to a second drive > > failure. > > > > Art S. Kagel > > > > Sent via Deja.com http://www.deja.com/ > Before you buy.
Art S. Kagel <kagel@bloomberg.net> wrote in message news:39942B62.B7AAB426@bloomberg.net... > NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 > > wyn441 wrote: > > > > RAID0 - Striping without Parity > > Striped Disk Array without Fault Tolerance - minimum 2 disks - > > very, very fast. Sucks if you don't have a good backup all the > > time. > > RAID1 - Mirroring and Duplexing - minimum 2 disks (duplexing > > needs 2 controllers). High tolerance for faults, slow but oh, > > so, reliable. > > Slow compared to what? Compared to singleton disks RAID1 is actually > slightly to 50% faster on reads due to reduced head contention because > most RAID1 systems can read from both drives in parallel, and about > equal to singleton drives for writes. > So, the data blocks being read from 2 disks are not compared on the fly ? And in case of partial media failure we'll have 2 different copies. Imagine that one of the disks fails completely then. Best regards, Sergey
Sergey Turin wrote: > > Art S. Kagel <kagel@bloomberg.net> wrote in message > news:39942B62.B7AAB426@bloomberg.net... > > NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 NO RAID5 > > > > wyn441 wrote: > > > > > > RAID0 - Striping without Parity > > > Striped Disk Array without Fault Tolerance - minimum 2 disks - > > > very, very fast. Sucks if you don't have a good backup all the > > > time. > > > RAID1 - Mirroring and Duplexing - minimum 2 disks (duplexing > > > needs 2 controllers). High tolerance for faults, slow but oh, > > > so, reliable. > > > > Slow compared to what? Compared to singleton disks RAID1 is actually > > slightly to 50% faster on reads due to reduced head contention because > > most RAID1 systems can read from both drives in parallel, and about > > equal to singleton drives for writes. > > > > So, the data blocks being read from 2 disks are not compared on the fly ? > And in case of partial media failure we'll have 2 different copies. Imagine > that > one of the disks fails completely then. OK, here's the deal. If a RAID10 array suffers from PMF (partial media failure) on one drive and that damaged block is read from that drive Informix will detect the problem and mark that chunk down so you will know to replace the flaky drive and recover it from its mirror. If it is not detected it will eventually fail and be replaced, still no problems. The ONLY problem arises if the 'ggod' drive fails before the flaky one, not likely but possible. In this case the data on the damaged block(s) is lost forever. HOWEVER, all of the non-damaged blocks are recoverable. THIS IS NOT TRUE in a RAID5 configuration! If an undamaged drive fails its data will be rebuilt using either good parity and bad data from the flaky drive, or bad parity from the flaky drive and good data from the others. Either way you now have two damaged disk blocks for every PMF affected block on the flaky drive each of which may represent up to 64 Informix pages instead of one damaged block in the case of the rare RAID10 scenario. Note that while the RAID10 lost data scenario is rather unlikely there are at least 4 good drives dependent on the flaky one in the RAID5 scenario so increased permanent data loss is 4 times more likely than in RAID10. Believe me I'm not talking theory here but bitter experience that Obnoxio and at least three other active DBAs here on CDi will confirm having experienced. None of us will EVER use RAID5 again! Art S. Kagel