August 28, 2006 at 11:29 am
Greetings All,
With a new server about to be implimented I am forced with deciding on the disk layout.
Some info:
This server is exclusive for record entry for a certain application and there already exists another server for reporting. So there will rarely be a random query/write. Merge replication will be used between the servers. The server will be utilizing a 21 spindle Fibre SCSI SAN.
The record is written first to one table as an XML(stored in a text field until we migrate to 2005). Then a XML service processes the records and in so doing spawns a number of threads writing each to different tables.
Seeking the best performance:
My original idea is to seperate the tables across mutliple small arrays to allow for the fastest synchronous write possible.
Or
Would it be best to lump them altogether (one single RAID5) even though 99% of all writes follow and identical pattern?
Thanks for any feedback,
-Z
August 28, 2006 at 12:36 pm
Raid 5 isn't particulary recommended if there is a lot of writing activity. Looks splitting it up would be ideal. Can't say more since the largest raid I have are 4-disks in 1+0.
Here is a summary of hardware tuning:
http://www.sql-server-performance.com/hardware_tuning.asp
August 28, 2006 at 2:27 pm
You're right about RAID10 being the way to go.
Thanks Jo
-Z
August 29, 2006 at 1:24 pm
Their will be additional complexity managing the tables over a pool of file groups; especialy from a recovery view point and RI. I've spead db objects over devices using natural object boundaries. Putting data on a device, the log on a different device, backups on the same device as the logs, or it's own device. Indexes can go to a device. The XML input streams on a device diffent from the database data and log devices and the extract on a device.
Viewing 4 posts - 1 through 3 (of 3 total)
You must be logged in to reply to this topic. Login to reply