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 |
|
swoodring
Starting Member
5 Posts |
 Posted - 2002-06-24 : 09:14:40
|
| We are upgrading from sql 7.0 to sql 2000. We have a brand new server in which we are installing it on. We are trying to decide on the best setup for it. The db are a datawarehouse. So mainly used for input. Gets updated with new data once a day. We were going to have:Partician 1 : Raid 1 - OS and ApplicationsNow we are trying to decide on the other particians:We have 6 36 Gig Drives to work with. One of them is going to be a hot swappable So basically we have 5 36 Gig drives to work with. I've read that its better to divide up the data/logs/tempdb. Would the following setup be good for performance or if not what would you recommend: We were thinking of making all of them Raid 5. My network guys say that with Raid 5 I'll end up with about 70 % so that accounts to about 127 Gig. So I was thinking the following: partition 2 - Data Files 112 gigPartition 3 - Logs 10 GigPartition 4 - Tempdb 5 gigAny suggestion would be greatly appreciated - trying to get this set up with the best possible scenerio for performance. This is my first time setting up the server myself so I'd appreciate any inputStacy |
|
|
Page47
Master Smack Fu Yak Hacker
2878 Posts |
Posted - 2002-06-24 : 11:17:18
|
| The big question to ask yourself if read/write ratio. RAID 5 is typically a very poor (yet overused) choice for databases that do any writes. A write to a RAID 5 actually causes four physical I/Os to occur inorder to calculate and write the parity bit. Because of this, for non-read-only dbs, a RAID 5 is a very poor choice. A mirror (RAID 1) or striped-mirror (RAID 10) is substantially better.With 6 spindles, I would probably consider a 2-disk RAID 1 (36 gigs)for my log file for the dw. With the other 4 disks I would set up a RAID 10 (64 gigs) for the dw data. This doesn't address the system databases (master,model,tempdb)...I suppose I'd throw data and logs for all the system dbs on the 4 disk RAID 10 (maybe the log for tempdb could share the RAID 1 with the DW log. The reason a RAID 1 is great for the logs is that (except on ROLLBACK) all writes to the log are sequential. If you can isolate the dw log, you will have 1 disk contention and minimal latency as the read/write head will always be writing to the next sector. If you actually need 112 gigs for your dw, the RAID 10 isn't gonna work. You either need to buy more disks, or go with a non-mirrored RAID (5). You should note that performance will be less. Disks are cheap considering the cost of performance....usually...The Microsoft Press Performance Tuning Technical Reference has a good chapter on RAID configuration.<O> |
 |
|
|
royv
Constraint Violating Yak Guru
455 Posts |
Posted - 2002-06-24 : 11:32:21
|
| I agree with Page47 on the RAID 5 implementation. I am also speaking from experience unforuneatly. RAID 5 is poor for writing, so based on your requirements, go with something else.*************************Someone done told you wrong! |
|
|
|
|
|
|
server upgrade and configuration