Hi,

The SOAP server delivered by HP NED takes a huge amount of cpu cycles
during marshalling and de-marshalling the request. I tried to tune it
with some caching options but that is not really giving me much
mileage. Would you kindly point me some doccument/whitepaper on this
topic if you came across any. Thanks in advance.

On Dec 23, 1:14=A0am, Anupam Das <das.anupa...@gmail.com> wrote:
> Hi,
>
> The SOAP server delivered by HP NED takes a huge amount of cpu cycles
> during marshalling and de-marshalling the request. I tried to tune it
> with some caching options but that is not really giving me much
> mileage. Would you kindly point me some doccument/whitepaper on this
> topic if you came across any. Thanks in advance.

Anupam,
  Marshalling and de-marshalling are very CPU consuming operations and
I am not sure how much performance gain you can have by trying various
parameters. It will need faster CPU's. I have seen few people
allocating CPU's exclusively for this in order to avoid performance of
other applications. We use NuWave software for SOAP messaging and in
order to avoid the problem we ended up resorting to a dirty solution.
The solution was to write a program which reads DDL dictionary and
converts the DDL into a big string (you need to handle COMP fields if
you have). This new DDL was used by NuWave so that minimal XML
processing is done on Tandem. On the consumer side, we had to write
complex XSLT's to convert the messages back to actual fields. So the
load was migrated to an XML appliance.
Regards
Robo

On Dec 25, 11:23=A0pm, Robo <mybase2008-downl...@yahoo.com> wrote:
> On Dec 23, 1:14=A0am, Anupam Das <das.anupa...@gmail.com> wrote:
>
> > Hi,
>
> > The SOAP server delivered by HP NED takes a huge amount of cpu cycles
> > during marshalling and de-marshalling the request. I tried to tune it
> > with some caching options but that is not really giving me much
> > mileage. Would you kindly point me some doccument/whitepaper on this
> > topic if you came across any. Thanks in advance.
>
> Anupam,
> =A0 Marshalling and de-marshalling are very CPU consuming operations and
> I am not sure how much performance gain you can have by trying various
> parameters. It will need faster CPU's. I have seen few people
> allocating CPU's exclusively for this in order to avoid performance of
> other applications. We use NuWave software for SOAP messaging and in
> order to avoid the problem we ended up resorting to a dirty solution.
> The solution was to write a program which reads DDL dictionary and
> converts the DDL into a big string (you need to handle COMP fields if
> you have). This new DDL was used by NuWave so that minimal XML
> processing is done on Tandem. On the consumer side, we had to write
> complex XSLT's to convert the messages back to actual fields. So the
> load was migrated to an XML appliance.
> Regards
> Robo

Thanks Robo. This is exactly the answer I was looking for. I was to
ensure I didn't miss any performance parameter for marshalling. A few
days back, I happened to meet one local HP NonStop sales executive and
he asks me something like this - did you tuned the server to Skip
known-XML-tag processing or something like that. I was not sure what
is he talking about. I gone back to the manual but found nothing to
tune inline with what he said. So, I dropped a note in this forum to
get some inputs.

One more interesting thing I observed. If you look back the ITUG
presentation since 2003 or 2004, you may encounter several places
where SOA and NonStop SOAP server has been described. But you won't
find a single slide on performance benchmarking for the NonStop SOAP
server.

I am sure everyone of us understand well the pros and cos of adapting
SOA architecture in our design but HP should atleast highlight the
limitations well in advance so that customers are alert and they
understand what they are going to get.

Anupam Das wrote:
> On Dec 25, 11:23 pm, Robo <mybase2008-downl...@yahoo.com> wrote:
> 
>>On Dec 23, 1:14 am, Anupam Das <das.anupa...@gmail.com> wrote:
>>
>>
>>>Hi,
>>
>>>The SOAP server delivered by HP NED takes a huge amount of cpu cycles
>>>during marshalling and de-marshalling the request. I tried to tune it
>>>with some caching options but that is not really giving me much
>>>mileage. Would you kindly point me some doccument/whitepaper on this
>>>topic if you came across any. Thanks in advance.
>>
>>Anupam,
>>  Marshalling and de-marshalling are very CPU consuming operations and
>>I am not sure how much performance gain you can have by trying various
>>parameters. It will need faster CPU's. I have seen few people
>>allocating CPU's exclusively for this in order to avoid performance of
>>other applications. We use NuWave software for SOAP messaging and in
>>order to avoid the problem we ended up resorting to a dirty solution.
>>The solution was to write a program which reads DDL dictionary and
>>converts the DDL into a big string (you need to handle COMP fields if
>>you have). This new DDL was used by NuWave so that minimal XML
>>processing is done on Tandem. On the consumer side, we had to write
>>complex XSLT's to convert the messages back to actual fields. So the
>>load was migrated to an XML appliance.
>>Regards
>>Robo
> 
> 
> Thanks Robo. This is exactly the answer I was looking for. I was to
> ensure I didn't miss any performance parameter for marshalling. A few
> days back, I happened to meet one local HP NonStop sales executive and
> he asks me something like this - did you tuned the server to Skip
> known-XML-tag processing or something like that. I was not sure what
> is he talking about. I gone back to the manual but found nothing to
> tune inline with what he said. So, I dropped a note in this forum to
> get some inputs.
> 
> One more interesting thing I observed. If you look back the ITUG
> presentation since 2003 or 2004, you may encounter several places
> where SOA and NonStop SOAP server has been described. But you won't
> find a single slide on performance benchmarking for the NonStop SOAP
> server.
> 
> I am sure everyone of us understand well the pros and cos of adapting
> SOA architecture in our design but HP should atleast highlight the
> limitations well in advance so that customers are alert and they
> understand what they are going to get.

Is this problem of high CPU consumption for marshalling and de-marshalling something peculiar to the HP SOAP server?  

Is HP's SOAP server implemented particularly inefficiently, or is marshalling and de-marshalling very expensive in CPU cycles for SOAP servers on other systems, too?

On Dec 28, 1:57=A0am, Keith Dick <KeithD...@earthlink.net> wrote:
> Anupam Das wrote:
> > On Dec 25, 11:23 pm, Robo <mybase2008-downl...@yahoo.com> wrote:
>
> >>On Dec 23, 1:14 am, Anupam Das <das.anupa...@gmail.com> wrote:
>
> >>>Hi,
>
> >>>The SOAP server delivered by HP NED takes a huge amount of cpu cycles
> >>>during marshalling and de-marshalling the request. I tried to tune it
> >>>with some caching options but that is not really giving me much
> >>>mileage. Would you kindly point me some doccument/whitepaper on this
> >>>topic if you came across any. Thanks in advance.
>
> >>Anupam,
> >> =A0Marshalling and de-marshalling are very CPU consuming operations an=
d
> >>I am not sure how much performance gain you can have by trying various
> >>parameters. It will need faster CPU's. I have seen few people
> >>allocating CPU's exclusively for this in order to avoid performance of
> >>other applications. We use NuWave software for SOAP messaging and in
> >>order to avoid the problem we ended up resorting to a dirty solution.
> >>The solution was to write a program which reads DDL dictionary and
> >>converts the DDL into a big string (you need to handle COMP fields if
> >>you have). This new DDL was used by NuWave so that minimal XML
> >>processing is done on Tandem. On the consumer side, we had to write
> >>complex XSLT's to convert the messages back to actual fields. So the
> >>load was migrated to an XML appliance.
> >>Regards
> >>Robo
>
> > Thanks Robo. This is exactly the answer I was looking for. I was to
> > ensure I didn't miss any performance parameter for marshalling. A few
> > days back, I happened to meet one local HP NonStop sales executive and
> > he asks me something like this - did you tuned the server to Skip
> > known-XML-tag processing or something like that. I was not sure what
> > is he talking about. I gone back to the manual but found nothing to
> > tune inline with what he said. So, I dropped a note in this forum to
> > get some inputs.
>
> > One more interesting thing I observed. If you look back the ITUG
> > presentation since 2003 or 2004, you may encounter several places
> > where SOA and NonStop SOAP server has been described. But you won't
> > find a single slide on performance benchmarking for the NonStop SOAP
> > server.
>
> > I am sure everyone of us understand well the pros and cos of adapting
> > SOA architecture in our design but HP should atleast highlight the
> > limitations well in advance so that customers are alert and they
> > understand what they are going to get.
>
> Is this problem of high CPU consumption for marshalling and de-marshallin=
g something peculiar to the HP SOAP server? =A0
>
> Is HP's SOAP server implemented particularly inefficiently, or is marshal=
ling and de-marshalling very expensive in CPU cycles for SOAP servers on ot=
her systems, too?- Hide quoted text -
>
> - Show quoted text -

This problem has nothing to do with NonStop CPU's. Its the nature of
the operation. No matter how efficient algorithm they use, so many
tags and envelop has to be added and removed.
However when you compare it with a non Tandem platform, most of the
time those platforms are dedicated to perfom these kind of tasks and
don't have to compete with other applications.
We use Oracle Service Bus which couldn't copeup with the load. Finally
we started using an appliance called Datapower for all our dirty
XSLT's.
Since OSB or Datapower are almost dedicated to perform this task, we
just keep adding a domain or managed servers when we see a performance
degradation.
If you can't afford to sepnd on something like datapower, you will
need to spare few CPU's to perform this task (of course depending on
your DDL complexity).

On Dec 28 2009, 12:00=A0am, Anupam Das <das.anupa...@gmail.com> wrote:
> On Dec 25, 11:23=A0pm, Robo <mybase2008-downl...@yahoo.com> wrote:
>
>
>
>
>
> > On Dec 23, 1:14=A0am, Anupam Das <das.anupa...@gmail.com> wrote:
>
> > > Hi,
>
> > > The SOAP server delivered by HP NED takes a huge amount of cpu cycles
> > > during marshalling and de-marshalling the request. I tried to tune it
> > > with some caching options but that is not really giving me much
> > > mileage. Would you kindly point me some doccument/whitepaper on this
> > > topic if you came across any. Thanks in advance.
>
> > Anupam,
> > =A0 Marshalling and de-marshalling are very CPU consuming operations an=
d
> > I am not sure how much performance gain you can have by trying various
> > parameters. It will need faster CPU's. I have seen few people
> > allocating CPU's exclusively for this in order to avoid performance of
> > other applications. We use NuWave software for SOAP messaging and in
> > order to avoid the problem we ended up resorting to a dirty solution.
> > The solution was to write a program which reads DDL dictionary and
> > converts the DDL into a big string (you need to handle COMP fields if
> > you have). This new DDL was used by NuWave so that minimal XML
> > processing is done on Tandem. On the consumer side, we had to write
> > complex XSLT's to convert the messages back to actual fields. So the
> > load was migrated to an XML appliance.
> > Regards
> > Robo
>
> Thanks Robo. This is exactly the answer I was looking for. I was to
> ensure I didn't miss any performance parameter for marshalling. A few
> days back, I happened to meet one local HP NonStop sales executive and
> he asks me something like this - did you tuned the server to Skip
> known-XML-tag processing or something like that. I was not sure what
> is he talking about. I gone back to the manual but found nothing to
> tune inline with what he said. So, I dropped a note in this forum to
> get some inputs.
>
> One more interesting thing I observed. If you look back the ITUG
> presentation since 2003 or 2004, you may encounter several places
> where SOA and NonStop SOAP server has been described. But you won't
> find a single slide on performance benchmarking for the NonStop SOAP
> server.
>
> I am sure everyone of us understand well the pros and cos of adapting
> SOA architecture in our design but HP should atleast highlight the
> limitations well in advance so that customers are alert and they
> understand what they are going to get.- Hide quoted text -
>
> - Show quoted text -

Hi all,

I've done a fair amount of R&D work with NSSOAP V3 on a S system
albeit I have not paid any attention to CPU usage whilst a client is
firing web request into the Tandem.  Much of my research involved
coding user exits.  Also, given that I'm working with version 3 of
NSSOAP, I can only speak to the web service "server" side of things.
NSSOAP V3 does not support client w/s.  Coming in version 4, though.

If you look at the UserExit stuff (NonStop SOAP User's Manual) you
will see a user exit function called pre_service(). It receives a
parameter (object) of class "ServiceEnv".  This class has a method /
property called "skipParse".  If you set this to TRUE the XML document
is not parsed / validated.  However, you must parse the buffer
yourself in the pre_service() user exit, i.e., to build your PATHSEND
request structure.

Maybe this is what your Tandem connect was referring to; however, it
is definitetly not as simple as setting a parameter in the
nssoap.config file etc as implied by the sales rep.

Cheers,
David

On Jan 1, 10:33=A0am, wiseass <david.wiseman...@gmail.com> wrote:
> On Dec 28 2009, 12:00=A0am, Anupam Das <das.anupa...@gmail.com> wrote:
>
>
>
>
>
> > On Dec 25, 11:23=A0pm, Robo <mybase2008-downl...@yahoo.com> wrote:
>
> > > On Dec 23, 1:14=A0am, Anupam Das <das.anupa...@gmail.com> wrote:
>
> > > > Hi,
>
> > > > The SOAP server delivered by HP NED takes a huge amount of cpu cycl=
es
> > > > during marshalling and de-marshalling the request. I tried to tune =
it
> > > > with some caching options but that is not really giving me much
> > > > mileage. Would you kindly point me some doccument/whitepaper on thi=
s
> > > > topic if you came across any. Thanks in advance.
>
> > > Anupam,
> > > =A0 Marshalling and de-marshalling are very CPU consuming operations =
and
> > > I am not sure how much performance gain you can have by trying variou=
s
> > > parameters. It will need faster CPU's. I have seen few people
> > > allocating CPU's exclusively for this in order to avoid performance o=
f
> > > other applications. We use NuWave software for SOAP messaging and in
> > > order to avoid the problem we ended up resorting to a dirty solution.
> > > The solution was to write a program which reads DDL dictionary and
> > > converts the DDL into a big string (you need to handle COMP fields if
> > > you have). This new DDL was used by NuWave so that minimal XML
> > > processing is done on Tandem. On the consumer side, we had to write
> > > complex XSLT's to convert the messages back to actual fields. So the
> > > load was migrated to an XML appliance.
> > > Regards
> > > Robo
>
> > Thanks Robo. This is exactly the answer I was looking for. I was to
> > ensure I didn't miss any performance parameter for marshalling. A few
> > days back, I happened to meet one local HP NonStop sales executive and
> > he asks me something like this - did you tuned the server to Skip
> > known-XML-tag processing or something like that. I was not sure what
> > is he talking about. I gone back to the manual but found nothing to
> > tune inline with what he said. So, I dropped a note in this forum to
> > get some inputs.
>
> > One more interesting thing I observed. If you look back the ITUG
> > presentation since 2003 or 2004, you may encounter several places
> > where SOA and NonStop SOAP server has been described. But you won't
> > find a single slide on performance benchmarking for the NonStop SOAP
> > server.
>
> > I am sure everyone of us understand well the pros and cos of adapting
> > SOA architecture in our design but HP should atleast highlight the
> > limitations well in advance so that customers are alert and they
> > understand what they are going to get.- Hide quoted text -
>
> > - Show quoted text -
>
> Hi all,
>
> I've done a fair amount of R&D work with NSSOAP V3 on a S system
> albeit I have not paid any attention to CPU usage whilst a client is
> firing web request into the Tandem. =A0Much of my research involved
> coding user exits. =A0Also, given that I'm working with version 3 of
> NSSOAP, I can only speak to the web service "server" side of things.
> NSSOAP V3 does not support client w/s. =A0Coming in version 4, though.
>
> If you look at the UserExit stuff (NonStop SOAP User's Manual) you
> will see a user exit function called pre_service(). It receives a
> parameter (object) of class "ServiceEnv". =A0This class has a method /
> property called "skipParse". =A0If you set this to TRUE the XML document
> is not parsed / validated. =A0However, you must parse the buffer
> yourself in the pre_service() user exit, i.e., to build your PATHSEND
> request structure.
>
> Maybe this is what your Tandem connect was referring to; however, it
> is definitetly not as simple as setting a parameter in the
> nssoap.config file etc as implied by the sales rep.
>
> Cheers,
> David- Hide quoted text -
>
> - Show quoted text -

Correction on previous posting... replace first reference to
"pre_service()" with "pre_process()", i.e., should read, "...will see
a user exit function called pre_process()."

David