This content has been marked as final. Show 8 replies
We run an 22.214.171.124 OLTP system on an EMC SAN with FAST Cache storage tiers. The FAST Cache software automatically migrates hot and cold blocks to/from the SDD drives. So far it seems to work real well and requires little administration on our end to keep track of what should and shouldn't be on the SDD drives.
What about the performance on writting?SDDs are different. They may differ drastically in terms of speeds in different use patterns.
Interfaces of SSDs are different too. Most have SATA III. Some have PCIe (the fastest ones). How fast is RAID 5 it depends on a SATA controller, most often it is so called "software RAID". In case of PCIe interface, RAID is totally "software". It will use CPU to calculate checksums and other RAID 5 data.
What about the lifecycle of SSDs that are quite less than HDDs? I've heard from the vender that its about 6 years.lifecycle of SSDs is not measured in time. It is measured in writes.
SSDs have a limit on a number of rewrites per block. Which is usually 1M. Time depends in intensity of re(writing) in your case.
It means, in general, the bigger SSD the longer it can last.
KeenOnOracle wrote:With RAID5 for EVERY data block written, the XOR block must be rewritten;
I'm running oracle 126.96.36.199.0 on windows environment.
We need to change our server that has began too old to attend our needs
We're planning to run a SSD raid 5 for improve performance.
which is a measurable penalty for write intensive objects like REDO files.
In my opinion, it is still a bleeding-edge technology. For my purposes, I avoid bleeding edge. Your purposes may be different. But you see, asking who's using it in production doesn't get much actual response. You have some amount of control over how interesting are the times you live in. The danger lies in making decisions based on what the vendors are selling, rather than your database needs. I can't see that much changes over a year, certainly in not how Oracle writes redo (although it has changed for exadata, the question becomes how long until the ability to [url http://structureddata.org/2011/10/12/exadata-smart-flash-logging-explained/]parallel write and only care about the fastest device is allowed to the rest of the world). I think it is hard to predict whether the self-tuning aspects of Oracle will handle these oddly fast devices appropriately, and if there are issues there, there could be some lag time before enough people complain that Oracle deals with it. Of course, they may be way ahead of me there, they aren't going to be completely reactive with a new version coming out, perhaps the developers got to play with new toys.
The EMC device someone mentioned sounds pretty cool, although I still have bitter memories of HP Autoraid moving stuff around inappropriately.
Of course the space shuttle was cool too. Note that the US is now dependent on big dumb Russian rockets.
I think it is hard to predict whether the self-tuning aspects of Oracle will handle these oddly fast devices appropriately, and if there are issues there, there could be some lag time before enough people complain that Oracle deals with it. Of course, they may be way ahead of me there, they aren't going to be completely reactive with a new version coming out, perhaps the developers got to play with new toys.Just for information. Oracle offers whole line of flash storage devices, from SSDs to Flash Array.
they are not as fast as others (like OCZ Z-Drive R4), but
these devices are certified with Oracle Database 11g R2 Smart Flash Cache - kind of transparent extension of buffer cache to flash memory (SSD).