SQL Server Performance Forum – Threads Archive
Which XEON?
I’ve been trying to find information on the benefits of one type of XEON over another and wonder if anyone else has any links/advice. I have the option to get either; 2 x 3GHz XEON 512kb cache or2 x 2GHz XEON MP 2MB cache for a W2K SQL2000 Enterprise Any ideas which is going to be best for a SQL 2000 datawarehouse?
ta Justin
I would go for second one 2x2Ghz with 2M cache. _________
Satya SKJ
I would go with the 2x2GHz as well. Try out hyperthreading too…
assuming that you are set on 2 processors and do not intend on going to 4, i would recommend the 2x3GHz as being better for datawarehouse applications,
the large cache is better for OLTP apps,
i would also actually try to get the 3.06GHz (533MHz FSB) i would recommend caution on the hyperthreading, don’t assume it will improve performance unless you actually test it, it is possible that it may have negative impact on large select queries typical of DW apps, insert performance may improve, but i don’t have complete study on this just yet there is some data on the matter of frequency versus cache in my article:
http://www.sql-server-performance.com/jc_sql_server_quantative_analysis1.asp but i do not draw the explicit conclusions
I would choose the 2 MB cache CPU. The two CPU specs taken in isolation don’t offer much in terms of choice. For your Data Warehouse ,a lot of cache is important, so is the size of the data and the types of DSS queries. For small and medium-scale DW scenarios, I don’t seriously think there would be a timesaving difference. A lot of RAM and Disk space would you a lot of good also. On hyperthreading, SQL Server and other issues try looking at this document: http://www.infoworld.com/article/02/02/25/020225plxeon_1.html]
NHO
in a DW app, i am assuming you will have hash and merge joins involving many rows
it is one thing to imagine what a big cache might do, it is another to actually work it out, below are the approximate cost formulas for hash & merge joins on a PIII600/256K versus PIIIXeon700/2M hash:
PIII 600MHz/256K : 520K + 8.5K per row
PIII X 700MHz/ 254K + 8.5K per row merge
PIII 600MHz/256K : 150K + 6.5K per row
PIII X 700MHz/ 54K + 6.5K per row my expectation is that Xeon 3.0GHz/512K and 2.0GHz/2M will show something similar in an OLTP app involving small rount count ops, the big cache clearly wins in DW with queries in 1M+ rows, you can do the math yourself
]]>