900 Millisecond Performance Spikes
We have a Forte 3.0.E.0 application running on an NT 4.0 server with an
Oracle 7.3.3 database, the following performance problem is being
observed. On a random basis, the time required to execute various
methods takes an additional 900 milliseconds. For example, a method
that repeatedly executes in 11 milliseconds peaks at 911 milliseconds
with or without other processes competing for processor time. This
performance peak happens with some regularity.
These peaks have been observed in all areas of the application. We have
taken steps to try to isolate the problem. Stop watches have been
placed in methods to gather data. A series of tests involved ending a
non-Forte service or process, executing a method and gathering data.
This was repeated until all non-Forte or non-essential NT services were
shutdown. No database access was performed by the monitored Forte
method. The Oracle server was stopped as part of the test. The problem
remained though the frequency of its occurrence diminished. Also, using
NT's performance monitor, no related performance hits occurred during
the observed Forte peaks. Similar results were observed within both the
development environment and deployed code. We have run this application
locally to eliminate network issues.
In an attempt to isolate the problem to Forte, a similar application was
written using Delphi, executed on the server and no performance peaks
were observed.
The server is a 166mhz single processor IBM Server 704 and has 256MB of
memory.
We would greatly appreciate any thoughts or suggestions you may have.
Rob Kovitz
Goodyear Tire & Rubber Co.
usgtrvqv@ibmmail.com
(330)796-5360
[1836 byte] By [
] at [2007-11-25 5:02:10]
>-
>Von:Hoffmann, Jens Michael
>Gesendet:Montag, 16. Februar 1998 09:53
>An:'Rob Kovitz'
>Betreff:AW: 900 Millisecond Performance Spikes
>
>Hi Rob,
>
>We have a Forte 3.0.E.0 application running on an NT 4.0 server with an
>Oracle 7.3.3 database, the following performance problem is being
>observed. On a random basis, the time required to execute various
>methods takes an additional 900 milliseconds.
>
>Unfortunately the millisecond timing feature of the StopWatch class (if you
>are using it) is
>broken (at least in 3.0.E.0) -- works only for seconds.
>
>
>
>Jens-Michael Hoffmann (jensmh@gmx.net)
>PARCS IT-Consulting GmbH
>
>
at 2007-6-29 9:22:45 >
Could this not be the garbage collector that kicks in? Just a thought...
>--Original Message--
>From:Sat Bardi [SMTP:satb@infogain.com]
>Sent:Monday, February 16, 1998 11:15 AM
>To: 'Rob Kovitz'
>Cc: 'kamranamin@yahoo.com'
>Subject:RE: 900 Millisecond Performance Spikes
>
>Hi,
>We have also seen a similar problem with a method call taking an additional
>900 milliseconds. We were running with forte 3.0.E.0 on NT 4.0 using a
>Pentium Pro 200 with 96MB Ram. However, we were not connecting to a database
>at the time nor were we using Oracle with the application.
>Hope this helps to narrow down your search.
>
>Unfortunately we haven't had the time to look into this further yet. I will
>email you if we do find something.
>
>
>Sat Bardi
>Infogain (UK).
>satb@infogain.com
>sat@go5.com
>
>
>--Original Message--
>From:Rob Kovitz [SMTP:kovitz@stratos.net]
>Sent:Sunday, February 15, 1998 8:24 PM
>To: kamranamin@yahoo.com
>Subject:900 Millisecond Performance Spikes
>
>We have a Forte 3.0.E.0 application running on an NT 4.0 server with an
>Oracle 7.3.3 database, the following performance problem is being
>observed. On a random basis, the time required to execute various
>methods takes an additional 900 milliseconds. For example, a method
>that repeatedly executes in 11 milliseconds peaks at 911 milliseconds
>with or without other processes competing for processor time. This
>performance peak happens with some regularity.
>
>These peaks have been observed in all areas of the application. We have
>taken steps to try to isolate the problem. Stop watches have been
>placed in methods to gather data. A series of tests involved ending a
>non-Forte service or process, executing a method and gathering data.
>This was repeated until all non-Forte or non-essential NT services were
>shutdown. No database access was performed by the monitored Forte
>method. The Oracle server was stopped as part of the test. The problem
>remained though the frequency of its occurrence diminished. Also, using
>NT's performance monitor, no related performance hits occurred during
>the observed Forte peaks. Similar results were observed within both the
>development environment and deployed code. We have run this application
>locally to eliminate network issues.
>
>In an attempt to isolate the problem to Forte, a similar application was
>written using Delphi, executed on the server and no performance peaks
>were observed.
>
>The server is a 166mhz single processor IBM Server 704 and has 256MB of
>memory.
>
>We would greatly appreciate any thoughts or suggestions you may have.
>
>Rob Kovitz
>Goodyear Tire & Rubber Co.
>usgtrvqv@ibmmail.com
>(330)796-5360
>
at 2007-6-29 9:22:45 >
It seems to me that this peaks are related to the garbage collector of
Forte.
You can check this by setting following log flags :
trc:os:1:1 : logs garbage collection activities
trc:os:5:5 : used in conjunction with trc:os:1:1 will show all collected
types and sizes on stable collects
There are also some TechNotes on the Forte site regarding this issue.
One of them explains how to read the trace generated by trc:os:5:5.
You can monitor your memory usage also by looking at the instruments of
the OperatingSystem Agent.
I hope this helps,
Steven Arijs
Rob Kovitz wrote:
>
> We have a Forte 3.0.E.0 application running on an NT 4.0 server with an
> Oracle 7.3.3 database, the following performance problem is being
> observed. On a random basis, the time required to execute various
> methods takes an additional 900 milliseconds. For example, a method
> that repeatedly executes in 11 milliseconds peaks at 911 milliseconds
> with or without other processes competing for processor time. This
> performance peak happens with some regularity.
>
> These peaks have been observed in all areas of the application. We have
> taken steps to try to isolate the problem. Stop watches have been
> placed in methods to gather data. A series of tests involved ending a
> non-Forte service or process, executing a method and gathering data.
> This was repeated until all non-Forte or non-essential NT services were
> shutdown. No database access was performed by the monitored Forte
> method. The Oracle server was stopped as part of the test. The problem
> remained though the frequency of its occurrence diminished. Also, using
> NT's performance monitor, no related performance hits occurred during
> the observed Forte peaks. Similar results were observed within both the
> development environment and deployed code. We have run this application
> locally to eliminate network issues.
>
> In an attempt to isolate the problem to Forte, a similar application was
> written using Delphi, executed on the server and no performance peaks
> were observed.
>
> The server is a 166mhz single processor IBM Server 704 and has 256MB of
> memory.
>
> We would greatly appreciate any thoughts or suggestions you may have.
>
> Rob Kovitz
> Goodyear Tire & Rubber Co.
> usgtrvqv@ibmmail.com
> (330)796-5360
at 2007-6-29 9:22:45 >
Hi,
We have also seen a similar problem with a method call taking an additional 900 milliseconds. We were running with forte 3.0.E.0 on NT 4.0 using a Pentium Pro 200 with 96MB Ram. However, we were not connecting to a database at the time nor were we using Oracle with the application.
Hope this helps to narrow down your search.
Unfortunately we haven't had the time to look into this further yet. I will email you if we do find something.
Sat Bardi
Infogain (UK).
satb@infogain.com
sat@go5.com
--Original Message--
From:Rob Kovitz [SMTP:kovitz@stratos.net]
Sent:Sunday, February 15, 1998 8:24 PM
To: kamranamin@yahoo.com
Subject:900 Millisecond Performance Spikes
We have a Forte 3.0.E.0 application running on an NT 4.0 server with an
Oracle 7.3.3 database, the following performance problem is being
observed. On a random basis, the time required to execute various
methods takes an additional 900 milliseconds. For example, a method
that repeatedly executes in 11 milliseconds peaks at 911 milliseconds
with or without other processes competing for processor time. This
performance peak happens with some regularity.
These peaks have been observed in all areas of the application. We have
taken steps to try to isolate the problem. Stop watches have been
placed in methods to gather data. A series of tests involved ending a
non-Forte service or process, executing a method and gathering data.
This was repeated until all non-Forte or non-essential NT services were
shutdown. No database access was performed by the monitored Forte
method. The Oracle server was stopped as part of the test. The problem
remained though the frequency of its occurrence diminished. Also, using
NT's performance monitor, no related performance hits occurred during
the observed Forte peaks. Similar results were observed within both the
development environment and deployed code. We have run this application
locally to eliminate network issues.
In an attempt to isolate the problem to Forte, a similar application was
written using Delphi, executed on the server and no performance peaks
were observed.
The server is a 166mhz single processor IBM Server 704 and has 256MB of
memory.
We would greatly appreciate any thoughts or suggestions you may have.
Rob Kovitz
Goodyear Tire & Rubber Co.
usgtrvqv@ibmmail.com
(330)796-5360
at 2007-6-29 9:22:45 >
