 | |
RAID Best Practices
sybase.public.sqlanywhere.general sybase.public.sqlanywhere.general
 |
| | Thread Tools | Display Modes |
#31
 
| |||
| |||
|
Re: RAID Best Practices -
08-06-2009
, 09:35 AM
|
"Markus KARG" <karg (AT) quipsy (DOT) de> wrote in message news:4a7ad15b$2 (AT) forums-3-dub (DOT) sybase.com... Actually you do have protection from hardware failure, since you have two replacements: The DB for broken LOG and MLG, the LOG for broken DB and MLG, and the MLG for broken DB and LOG. What you mean is protection from long restore times, I am assuming? Well actually I meant the fact that you are not using hardware to replicate any of the files. I would have thought that you'd want a configuration that: a) replicated via hardware b) mirrored via software They sort of protect against a wider variety of problems although we haven't had so many that I can distinguish between them. If you have hardware failure, the likelihood is that one of the dbs will be fine and the other one is unusable. If you have a software problem then the software will replicate over to the mirror. Having said that, it could still be argued that you can recover from the same situations with individual disks (maybe - haven't thought it through completely yet). I remember a situation: ENGINEER (to customer): Sorry, your database is corrupt and unusable. CUSTOMER: Not to worry. We use RAID and have a hardware backup. ENGINEER: I am sorry to inform you that you have 2 unusable databases!  Now, even if the software mirror is enough, why configure it that way and lose the hardware failsafe and performance improvements? It is unlikely you'd get equal i/o distribution across all four disks as separate disks. I assume you also do database unloads and reloads on occasion. Would you not be thrashing only one disk in that instance when this downtime could be almost twice as fast? |
academic. I can remember two customers having used hardware raid and ended
up with screwed information thanks to a bug in the controller's firmware.
Also, the initial discussion was not about software RAID vs. hardware RAID
but about putting DB and LOG on different disks (or RAID sets), or putting
both together on a RAID set. I do not see that the risk of software faults
is higher or lower in any of both cases.
Quote:
|
This is what I think, too. But as this estimation is rather complex, it would be great if the Sybase engineer that wrote into the manual that using multiple standalone disks is a good performance idea would post a statement here why not instead writing that RAID is an even better idea. I agree. I just know that whilst we don't know the exact performance/reliability improvements of each configuration, there are enough arguments weighted for the RAID 10 solution for holding all your database files such that its not worth spending too much time looking at alternatives. But certainly input from Sybase is always very welcome and informative. |
currently we are discussing technical *ASSUMPTIONS*. ;-)
A technical fact would be a direct comparison of the three suggested
solutions on exactly the same hardware with exactly the same load. Everthing
else is an academic hypothesis. ;-)
Regards
Markus
#32
| |||
| |||
|
|
A technical fact would be a direct comparison of the three suggested solutions on exactly the same hardware with exactly the same load. Everthing else is an academic hypothesis. ;-) Regards Markus |
approaches and your typical workload will give you the technical facts
you seem to seek. Everything else will be doomed to be based on the
classical "It depends"
Regards
Volker
#33
| |||
| |||
|
|
There are really only two files you need to worry about: the .db file and the log file. |
Breck
--
Breck Carter http://sqlanywhere.blogspot.com/
RisingRoad SQL Anywhere and MobiLink Professional Services
breck.carter (AT) risingroad (DOT) com
#34
| |||
| |||
|
|
Markus KARG wrote: A technical fact would be a direct comparison of the three suggested solutions on exactly the same hardware with exactly the same load. Everthing else is an academic hypothesis. ;-) Regards Markus My personal impression is that only your own tests with the different approaches and your typical workload will give you the technical facts you seem to seek. Everything else will be doomed to be based on the classical "It depends" Regards Volker |
#35
| |||
| |||
|
|
On 6 Aug 2009 04:53:43 -0700, ns_dkerber (AT) ns_WarrenRogersAssociates (DOT) com> wrote: There are really only two files you need to worry about: the .db file and the log file. The temporary file can be important, too, depending on the workload. |
into consideration that each write to LOG will impose a parallel write to
MLG if enabled...
Guys, believe me, I knew exactly what I initially asked and why I actually
asked exactly that... ;-)
Regards
Markus
#36
| |||
| |||
|
|
Today the typical server has lots of HDDs, mostly configured as RAID sets. The SA11 manual says that for best performance, the DB and the LOG file shall be on different drives. This offers a lot of options, and the question is, which is really optimal. When we imagine a server with four HDDs, these could be used in several ways: * RAID5 or 10 made up from all four disks, hosting anything (system, DB, LOG, MLG, TEMP). * Two Mirrors, one hosting system, DB and MLG, the other hosting LOG and TEMP. * Four independent disks, hosting system+DB, LOG, MLG, TEMP * etc. So before doing endless trials: Is there a best practice how to use all the disks? Is RAID better or worse than using each disk independ? Regards Markus |
#37
| |||
| |||
|
|
Hi Markus, So RAID10 is faster than RAID5? How much? Yes. RAID10 is faster than a single hard disk. RAID 5 is slower than a single hard disk. That should be enough! |
I recently made a lot of tests and I can say that an array raid5 with 3 hd
is two time faster than a single hard disk
Bye
Quote:
|
"Markus KARG" <karg (AT) quipsy (DOT) de> wrote in message news:4a76d154$4 (AT) forums-3-dub (DOT) sybase.com... Thomas, thank you for your comments. More inlined. - From my knowledge the best overall throughput for a DB is RAID 10. RAID 5 is to slow with all calculation. So RAID10 is faster than RAID5? How much? - I would never ever put a System Critical Database on a Single Hard Disk. Minimum RAID1 !!! But when you have DB on one drive and LOG on another drive you also have a mirror in some way since you can replace a broken DB by an old DB plus the existing LOG, and you can replace a broken LOG by an existing MLG. So the only benefit is to reduce downtime. We run our Database with DB and LOG from a external RAID10 Device. Temp on a LOCAL Raid1 Partition Why temp on a mirror instead of simple stripe set (RAID0)? To reduce downtimes? Regards Markus |
|
« Previous Thread
|
Next Thread »
| Thread Tools | |
| Display Modes | |
Linear Mode
| |
Powered by vBulletin Version 3.5.3
Copyright ©2000 - 2012, Jelsoft Enterprises Ltd.
Copyright ©2000 - 2012, Jelsoft Enterprises Ltd.