Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Tuxedo Ap Server crushing Oracle Server?

Re: Tuxedo Ap Server crushing Oracle Server?

From: Andrew Babb <andrewb_at_mail.com>
Date: Mon, 19 Apr 1999 07:46:23 +0800
Message-ID: <371A6ECF.FD3CCAF7@mail.com>


Hi Ron,

We successfully deployed a 7x24 mission critical system on Tuxedo. However, it did take time, and expert support from both BEA and Oracle. A couple of the problems that we created (and solved) are as follows;

We started of with the concept of the 'Mega Server' where every Tuxedo Service was defined within a single TUX Server, and then we ran about 20 copies of this single TUX Server. However, we ran into a couple of problems, namely the size of Oracle's Session Pool (used by TUX and MTS to store the session variables) and Oracle's Session Switching. Check Session Memory size (and Free Memory) from v$sgastats and Session Switching from v$latchstat.

After convincing the development teams that this was not working, a long exercise of breaking TUX Services into multiple small TUX Servers, and expanding the application design, we came up with a good grouping of Services to Servers and a concept we called 'Smart Binding' that meant if the required Service was in the current Server, we did not do a TUX call, but simply called locally within the Server (reducing TUX Service calls and switching). Having completed this, we ended up with 2 or 3 copies of each TUX Server, and the total number of Servers increased by between 50% and 100%.

We also ran numerous (I think in excess of 2 man years) Stress Testing before we deployed, and the Oracle init parameters undertook a number of changes, with test runs changing just one Oracle parameter at a time, like SPIN_COUNT moving from 8000 -> 10000.

What I am trying to say is that Oracle and Tux works, but you have to be prepared to put in the extra effort, and deploying Oracle & Tuxedo systems take longer than normal.

Just my 0.02
Andrew

Ron Denham wrote:

> My company just upgraded to PeopleSoft Financials 7.0 from PeopleSoft 5
> and we are trying to us the Tuxedo application server with PeopleSoft.
> We are a very established PeopleSoft shop with a very demanding user
> base.
>
> We currently have a HP K570 with 2GB RAM and 6 processors as the Oracle
> 7.3.4 database server and a HP K570 with 2GB RAM as the Tuxedo v6 middle
> tier app server. About 150 concurrent users use PeopleSoft (now coming
> in via 40 app server connections to the database) and we have another
> 100 to 120 that come into the same instance as part of other financial
> related systems.
>
> Prior to the upgrade, with all 270 (PS & non PS) users coming in 2-tier,
> the K570 box handled this fine with peaks of 70% utilization.
>
> After the upgrade, we now have 100% utilization on the database server
> all day long. My top 30 CPU users in the database are all Tuxedo
> connections. The system is very sluggish as a result.
>
> It looks like the database server is in panic mode trying to service the
> 40 app server connections, which all seem to be dropping requests to the
> database in parallel. Previously, we had more connections in the
> database but the connection requests were more serial in nature... so my
> theory is that the system formerly did not have to juggle so many
> parallel requests. Also my theory is that Tuxedo does a very good job
> of splitting up requests across the connections it establishes.
>
> My question is... Has anyone else seen this behavior? Is my only
> alternative to go back to 2-Tier mode or to buy a V Series HP box with
> 24 processors?
>
> Any advice would be appreciated.
>
> - Ron
>
> ronnied_at_mindspring.com
Received on Sun Apr 18 1999 - 18:46:23 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US