windows-1252?Q52=65=3A=42=75=66=66=65=72=20=61=6C=

There can be several reasons for the balance between data and index pages in the buffer cache to be 'abnormal' for want of a better word and some of these are perfectly valid and nothing to worry about. As to what does all this mean, it's as it seems, the percent of the buffer cache pages allocated to pages that contain data is the Data % at the bottom of the onstat -P report while the number of pages allocated to pages containing index nodes and leaves contribute to the Index %. Other is what's left over, ie bitmap and other overhead pages. One reason for things to get out of balance you need not be concerned about, that being a bug in early releases of IDS 7.31. IDS 9.40 uses a different buffer aging algorithm and 10 introduced an even more advanced algorithm, so that's no longer a problem. Valid reasons for imbalance: - Summary reports run in batch that rely on indexed queries will tend to flood the cache with index pages - Large batch data loads will tend to push out index pages in favor of data pages - History reports that tend to perform table scans will also skew the buffer cache contents towards data pages - Large concerted sessions of UPDATE STATISTICS will indeed push the percentage of index pages up Any of these situations may cause short lived performance problems at the beginning of the 'normal' processing day. Some sites run 'primer' applications that perform typical queries as a batch shortly before the beginning of normal operations to prime the buffer cache with important data. For example a bank may select the customer records for all customers who executed transactions at a particular branch the previous day on that branch's local ER node so the records are in cache if the customer returns. The other category is usually only high if the buffer cache is larger than needed and there are therefore many unused buffers (the count in the 'Other' column for partnum 0). Even so, this is most common shortly after the server has been brought online and disappears over time unless the cache is severely oversized. Art S. Kagel ----- Original Message ----- From: H.G <ids@iiug.org> To: ids@iiug.org At: 10/30 15:57:24 Greeting All, IDS 10.0.FC6, Linux AS4 Need some help in understanding the buffer allocated to data, Index and others from onstat -P | tail output. From my memory, in one of the article/script (probably by Art or may be other guru), it was suggested that in OLTP, data % should be > 90%. One of our PROD database servers occasionally shows data % as low as 30% and Index % as 60%. Sometimes other % appears to be 40%, Index % 50% and data 10%. Something like this. I have a feeling that when I see index % or other % higher than data %, server appears to have some performance issues. Can someone please point me to the right direction for? 1. What is buffer allocated to data, index and others? 2. What may be the cause of this? 3. Could it be related to update stats as update stats are being done by distribution only? 4. What ONCONFIG parameter(s) need an attention? Please let me know if I need to post any onstat output. Thanks. __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.