October 19, 2001 at 3:57 pm
I bought the SQL 2K book. It covers, as would be expected, some of the performance enhancements that are found in SQL 2K.
K. Brian Kelley
http://www.sqlservercentral.com/columnists/bkelley/
K. Brian Kelley
@kbriankelley
November 5, 2001 at 9:13 am
Andy,
I have seen statistics that indicate the performance of clustered server will increase as the queries are shared across multiple servers. I would be interested in knowing whether you find this to be true. I am considering this approach on a 4 way 1.5 gig box.
quote:
Chris,We're using a very expensive 3rd party app that makes extensive use of server side cursors. When we had it running on a 4 way 550 with 1g of RAM it was maxed out. Moving to 2g helped a lot, we were short on memory. Thats with about 200 concurrent users. Not much I can do to fix the app, so only choice was to buy more power. 8 way box plus an external fibre channel array (9 drives plus a hot spare) now runs at avg of 10% utilization. We needed room to grow. More importantly, if you're running near max the box becomes much more sensitive to load - which would kill response times.
I'd like to cluster at some point, but I think if/when we do it will be with another 8 way box. I think box plus drives, OS, cost us around 80k. Money well spent.
Andy
November 5, 2001 at 11:25 am
So far all my queries are running on one box. I would think that if you're using distributed partitioned views there might be some gains.
Andy
November 5, 2001 at 11:44 am
Clustered servers shouldn't give any help. Clustering is a fault tolerance, not a performance increase. The query executes on one box (still).
For partioned views, you get two types of benefits. One, the load for the queries can be distributed to more than one box (I query box A, you query box B). The other benefit is the servers "send" the part of the query to the other box, so they do not spend CPU cycles processing the query. Of course, the first box still has to assemble the results.
Steve Jones
November 7, 2001 at 7:33 am
I stand, er sit, corrected. My old and feable mind tends to get the tech terms mixed up. I meant partitioned views but we are moving to a fault tolerant clustered approach here. Ahh, I mind is a terrible thing to waste. I guess I'm reaping the wild oats I sowed in the 70's.
Thom
quote:
Clustered servers shouldn't give any help. Clustering is a fault tolerance, not a performance increase. The query executes on one box (still).For partioned views, you get two types of benefits. One, the load for the queries can be distributed to more than one box (I query box A, you query box B). The other benefit is the servers "send" the part of the query to the other box, so they do not spend CPU cycles processing the query. Of course, the first box still has to assemble the results.
Steve Jones
November 7, 2001 at 11:21 am
Viewing 6 posts - 16 through 20 (of 20 total)
You must be logged in to reply to this topic. Login to reply