Path: dp-news.maxwell.syr.edu!spool.maxwell.syr.edu!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!postnews.google.com!f14g2000cwb.googlegroups.com!not-for-mail From: bdbafh@gmail.com Newsgroups: comp.databases.oracle.server Subject: Re: Oracle IO tuning tips/practices on Windows Date: 2 Feb 2006 15:18:32 -0800 Organization: http://groups.google.com Lines: 37 Message-ID: <1138922312.807196.20760@f14g2000cwb.googlegroups.com> References: <1138842750.019797.128420@g44g2000cwa.googlegroups.com> NNTP-Posting-Host: 137.237.150.123 Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" X-Trace: posting.google.com 1138922320 26787 127.0.0.1 (2 Feb 2006 23:18:40 GMT) X-Complaints-To: groups-abuse@google.com NNTP-Posting-Date: Thu, 2 Feb 2006 23:18:40 +0000 (UTC) In-Reply-To: User-Agent: G2/0.2 X-HTTP-UserAgent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.7.12) Gecko/20050915 Firefox/1.0.7,gzip(gfe),gzip(gfe) Complaints-To: groups-abuse@google.com Injection-Info: f14g2000cwb.googlegroups.com; posting-host=137.237.150.123; posting-account=rcQZawwAAAALUCo-zZUVUQgj4l_8W6zb Xref: dp-news.maxwell.syr.edu comp.databases.oracle.server:260621 Have you considered provisioning the RAID volumes in groups of 4 disks as RAID 10, and creating multiple datafiles per tablespace, such that you would have 3 mount points, one per RAID volume (aka LUN). LUN 0 ID 0,1,2,3 LUN 1 ID 4,5,8,9 LUN 2 ID 10,11,12,13 Create one partition per LUN, for each filesystem blocksize. If all tablespaces have an 8 KB blocksize, one partition per LUN would do, formatted with an 8192 byte blocksize. As you have 2 block sizes, create 2 partitions per LUN, create a second partition per LUN with a 16384 blocksize. Now you have a setup where the datafile blocks, filesystem and RAID stripes align. You have increased overhead in terms of having to create sets of datafiles for each tablespace, so its not as simple as your SAME setup. What you would gain, is that for single block reads, a read request would only have to involve a single drive. This should provide higher concurrency and lower average response time than your existing configuration. Currently, with a 64 KB stripe size (over 6 disks) and an 8 KB block size, half of the blocks per stripe are divided over multiple hard drives (1,4,5,8 are not). By not aligning the database file blocks and the physical hard drives twice as many physical reads are required half the time to access a single block. That is a significant penalty. hth. -bdbafh