Comments on: Oracle hidden costs revealed, part 1 – Does a batch job run faster when executed locally?

By: truss -T, pstack & prun « Oracle

truss -T, pstack & prun « Oracle — Fri, 29 Apr 2011 20:18:33 +0000

[...] list out the stack of function calls leading to the suspended call. This is clearly demonstrated in this post by Tanel [...]

By: How to CANCEL a query running in another session? | Tanel Poder's blog: Core IT for Geeks and Pros

Thu, 18 Feb 2010 04:17:07 +0000

[...] http://blog.tanelpoder.com/2008/02/05/oracle-hidden-costs-revealed-part-1/ [...]

By: Bill

Bill — Wed, 30 Jul 2008 16:44:11 +0000

That would be great.
I have some DBAs very interested in this because we have some oracle report writer processes that are exhibiting these symptoms.
They are batch only programs but they seem to be spending a large amout of time spinning in calls to pollsys.
I am just a knukle dragging SA but when these processes run long you can bet who they scream at first.

By: tanelp

tanelp — Wed, 30 Jul 2008 16:38:20 +0000

Hi Bill,

I only have a draft with some test cases. I’ll see if I can write it up in coming week or so.

By: Bill

Bill — Wed, 30 Jul 2008 16:22:40 +0000

Can you tell me if you have published the second article you mentioned above?

By: tanelp

tanelp — Thu, 19 Jun 2008 11:50:56 +0000

Are you fetching lots of data? If returning lots of data to client, then local box access will always be faster (transfer data in server memory vs over Ethernet card, Switches, VLANs, WANs etc)

Which reporting tool are you using?

One possible reason for slowness is that you just have too small client arraysize (number of rows fetched at a time) and most of the extra time is wasted on network packet ping-pong between the client and server.

Run my snapper script on the session executing the report and post the output, then I’ll be able to help.

By: qihua

qihua — Thu, 19 Jun 2008 10:10:17 +0000

for a DW query, when running on local box, 1 hour will be took, but when run on remote box, 3 hours will be took, most of the time is waiting on “from client”, but the boxes are connected over LAN, and the speed is 1G, Don’t know why trasfer data over network by oracle net is so slow in comparison with the bandwidth of the netowrk

By: tanelp

tanelp — Fri, 23 May 2008 16:38:28 +0000

Thanks Ganesh,

On Linux Oracle did not have to do any polling, because:

1) it was still executing the same lotslios call
2) there potential client-sent breaks would have been detected using SIGURG signal handler when receiving an “out of band” packet with SIGURG flag, thus polling was not necessary

In other words, if there exists an OS level infrastructure for receiving out of band messages and notifying user processes about it, then no polling during call execution is needed

By: Ganesh

Ganesh — Tue, 20 May 2008 01:23:41 +0000

Hello:

Nice article.

One question though.

In the 27 odd seconds that it took for the linux client to run lotslios.sql, How come truss did not show even one call to pollsys()?

Ganesh

By: Bas

Bas — Sat, 16 Feb 2008 07:49:25 +0000

These kind of posts are great !