Please start any new threads on our new
site at https://forums.sqlteam.com. We've got lots of great SQL Server
experts to answer whatever question you can come up with.
| Author |
Topic |
|
MichaelP
Jedi Yak
2489 Posts |
 Posted - 2003-07-11 : 14:11:12
|
| I might be in the market for a high speed Storage subsystem. Right now, we've got everything (data, logs, indexes etc) on a RAID 5 array. Needless to say, this is really a problem as far as I/O goes.We don't have a ton of data (around 15-20GB's give or take) but we need to get at the data with a quickness. Does anyone have experience with some higher performing solutions like a SAN or NAS? I don't think NAS will help us much, because we already haev tons of network traffic on our gigabit network.I'm thinking a SAN is the way to go since we could use a single SAN for multiple database servers. We have one sql server setup in a failover cluster configuration, and others that are not. I think ideally we'd like to put all the data on this SAN so that we can access it very quickly.I read Wanderer's posts about the problems he was having with his SAN, so I'm a bit worried.Thoughs? Opinions? Any idea on the cost of such a system?Michael<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
tkizer
Almighty SQL Goddess
38200 Posts |
Posted - 2003-07-11 : 14:14:50
|
| We started using the SAN solution about a year ago on all of our production servers. It is super fast. We went with Compaq's solution which is EVA.Tara |
 |
|
|
setbasedisthetruepath
Used SQL Salesman
992 Posts |
Posted - 2003-07-11 : 14:44:25
|
| SAN's are very expensive.You might consider a RAID environment better than RAID 5, which is almost always the worst choice - RAID 10 or any of the redundant mirrored equivalents (RAID 20, 30, 50, etc).And did I read correctly that data and logs are on the same volume?Jonathan{0} |
|
|
|
MichaelP
Jedi Yak
2489 Posts |
Posted - 2003-07-11 : 16:22:09
|
| Jonathan,Yes, the TXlogs, Indexes, and Data for several databases are all on the same RAID 5 volume on a single SCSI channel. I pretty much thought that was was the worst possible configuration. I think we are limited on the number of arrays that we can use with our clustering hardware. I think if we can come up with a way to keep the the TX logs and TempDB's on a RAID 0 or RAID 10 array, that would really help things out.I think I read someplace that a SAN was up in the $250,000 range. Does that sound about right Tara?<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
|
setbasedisthetruepath
Used SQL Salesman
992 Posts |
Posted - 2003-07-11 : 16:49:49
|
| In that case you would find a significant performance boost by using a RAID 10 array for the data volume and a RAID 1 array for the transaction logs. You want mirroring on the tx logs versus naked striping b/c you're performing nearly 100% writes. I can't stress enough how bad it is to be doing sustained writes on a RAID 5 volume - the microcontroller must translate each write into 2 reads for parity computation in addition to the original write.For super high performance you might consider 2 RAID 10 arrays for data, with strategic placement of objects in each array through filegroups to maximize join and index retrieval throughput.For near-godlike performance you might consider 2 or more RAID 50 arrays for data as above. Though that costs beaucoup $$$$ since you're likely talking several UltraSCSI 320 controllers, probably multiplexed, each controlling 40+ drives. And at that stage you start hitting the upper limit of the PCI bus, so you have the two options of moving to a different bus architecture or a SAN.Edited by - setbasedisthetruepath on 07/11/2003 16:53:16 |
|
|
|
tkizer
Almighty SQL Goddess
38200 Posts |
Posted - 2003-07-11 : 17:04:05
|
quote:
I think I read someplace that a SAN was up in the $250,000 range. Does that sound about right Tara?
I'm not even sure.Tara |
|
|
|
eyechart
Master Smack Fu Yak Hacker
3575 Posts |
Posted - 2003-07-11 : 19:27:02
|
quote:
quote:
I think I read someplace that a SAN was up in the $250,000 range. Does that sound about right Tara?
I'm not even sure.Tara
That is about where you would start. You have to consider not only the disk hardware, but any software you might also add (powerpath, SRDF, etc. to use EMC examples). There is also the extra cost of FC switches (redundant hopefully) and HBAs for all systems that you are connecting. Then toss on maintenance...Very quickly the cost of these systems gets very high. -ecEdited by - eyechart on 07/11/2003 19:28:27 |
|
|
|
|
|
|
|
|
Fast Storage Subsystems