RE: Hints
From: MacGregor, Ian A. <ian_at_slac.stanford.edu>
Date: Tue, 16 Aug 2011 14:05:19 -0700
Message-ID: <FD1D618E4F164D4C8BA5513D4268174A015751BC9CC4_at_EXCHCLUSTER1-02.win.slac.stanford.edu>
I very much agree with what has been said, though some was a bit over the top. Here's the situation: A database is handling about 2800 connections per minute peak. These are not light-weight connections, but are pulling together information from on the fly user defined conic sections of the celestial sphere. The connections themselves come from perhaps 1000 client machines in a batch farm. The data is then processed via a pipeline controlled by the database which controls the order of the processing, and weeds out redundant steps.
From: Greg Rahn [greg_at_structureddata.org] Sent: Monday, August 15, 2011 1:53 PM
To: tim_at_evdbt.com
Cc: neil_chandler_at_hotmail.com; MacGregor, Ian A.; oracle-l_at_freelists.org Subject: Re: Hints
Date: Tue, 16 Aug 2011 14:05:19 -0700
Message-ID: <FD1D618E4F164D4C8BA5513D4268174A015751BC9CC4_at_EXCHCLUSTER1-02.win.slac.stanford.edu>
I very much agree with what has been said, though some was a bit over the top. Here's the situation: A database is handling about 2800 connections per minute peak. These are not light-weight connections, but are pulling together information from on the fly user defined conic sections of the celestial sphere. The connections themselves come from perhaps 1000 client machines in a batch farm. The data is then processed via a pipeline controlled by the database which controls the order of the processing, and weeds out redundant steps.
Shortly after this started one of the physicists had been ready one of Tim's old posts. He wanted to see what would happen if we the unmentionable parameter was changed. I advised against it, but finally thought that such a change would probably make the system worse. It did not, things improved dramatically. Oh we have a few queries where the buffer gets count are out-of-this-world. But overall the change turned out to be positive. I think the real problem was taking of the OS stats at a not so busy time. However the chances of reverting the parameter is not good.
From: Greg Rahn [greg_at_structureddata.org] Sent: Monday, August 15, 2011 1:53 PM
To: tim_at_evdbt.com
Cc: neil_chandler_at_hotmail.com; MacGregor, Ian A.; oracle-l_at_freelists.org Subject: Re: Hints
Or to put it another way: "make sure the scope of the solution matches the scope of the problem".
-- http://www.freelists.org/webpage/oracle-lReceived on Tue Aug 16 2011 - 16:05:19 CDT