March 9, 2009 at 6:43 am
I have a multi-server configuration for a BI system I am working on that has the following servers:
1. A server that contains ETL/Staging.
2. A server that contains the dimension and fact tables.
3. A server that contains the OLAP cubes
4. Initially 3 servers to handle the presentation using SharePoint and RS.
Each of the servers will have local disks for the OS and application binaries. The rest of the storage will utilize an HP SAN and therein lies the problem for me related to determine the number and size of the LUNs as I have never worked with a SAN. I have done quite a bit of research the last few days. Brian Knights book on SQL Server 2008 Administration has some good information on this which has helped. I have also been digesting the recent Implementing a SQL Server Fast Track Data Warehouse white paper on TechNet which also provided some useful information. There were also a couple of webcasts on the MSDN that were a few years old that provided some insight as well. I understand the basic principles of creating LUN's for specific functions/workloads such as data, logs and tempdb based on this research.
I know we should have separate LUNs for log, tempdb, and data.
The number one question I have is would you configure LUN's to be used by each server, or would you try to create LUNs' that would be shared across all the servers.
I realize there is some trial-n-error involved here that will need to be tested to confirm the performance.
March 9, 2009 at 11:41 am
I don't see why a server should access a LUN holding data it doesn't need. There's no performance or security emprovement in doing that. I would keep each LUN reserved to the server needing to access its data.
Regards
Gianluca
-- Gianluca Sartori
March 9, 2009 at 12:08 pm
Gianluca;
Thank you for your reply on this. I just finished reading the MS white paper Implementing a SQL Server Fast Track Data Warehouse (http://msdn.microsoft.com/en-us/library/dd459178.aspx#) which shows something where they use four LUNs for user data, tempdb and staging and then fifth LUN for logs. In addition the user data, tempdb and staging are spread across the LUNs.
Still trying to wrap my head around this, but it seems configuring the LUNs is more of an art than science and that whatever configuration you design needs to be rigorously tested to confirm the desired performance and then monitored to be sure it is maintained.
Dave
March 9, 2009 at 12:16 pm
Well, I'm no LUN expert, I can only tell you what I'm successfully doing with my configuration:
I have a 6 node cluster server with 9 instances. Each instance points to a data disk and a log disk, and each is a different LUN on a RAID 10 array. Every instance has no access to the disks belonging to the other instances.
I understand your problem is a bit different, but this is all I can say on this topic.
Gianluca
-- Gianluca Sartori
March 9, 2009 at 12:39 pm
Gianluca;
Thanks again. The MS white paper indcates their approach is the same as yours; the LUNs are dedicated to a server/instance.
Dave
Viewing 5 posts - 1 through 4 (of 4 total)
You must be logged in to reply to this topic. Login to reply