Eclipse Community Forums
Forum Search:

Search      Help    Register    Login    Home
Home » Eclipse Projects » Test and Performance Tools Platform (TPTP) » New Agent Questions
New Agent Questions [message #69577] Fri, 05 May 2006 20:32 Go to next message
Eclipse UserFriend
Originally posted by: user.domain.invalid

Hi all,

I need to write an agent that will run under hpux. I could write the
agent in java or c++ or c, doesn't matter. C++ would be the better
choice from the point of view that the app I have to test has a c++ api.
For that reason I think I should go with the new agent architecture.

However, if I do that, wouldn't it be true that I would have to run all
my tests headless because, if I read things right, none of the tptp
tools I've plugged into my Eclipse workbench at this point (tptp 4.1)
can communicate with the new agent controller?

So if I want to control my testing using the workbench, I have to write
my agent using the old agent architecture. Is that correct?

thanks,
Rick
Re: New Agent Questions [message #69597 is a reply to message #69577] Sat, 06 May 2006 21:56 Go to previous messageGo to next message
Randy D. Smith is currently offline Randy D. SmithFriend
Messages: 394
Registered: July 2009
Senior Member
rkm wrote:
> Hi all,
>
> I need to write an agent that will run under hpux. I could write the
> agent in java or c++ or c, doesn't matter. C++ would be the better
> choice from the point of view that the app I have to test has a c++ api.
> For that reason I think I should go with the new agent architecture.

So far so good. I agree with your analysis. The one glitch in your
thinking is the HP/UX portion... so far the new AC is only ported to
Intel architectures, though more are in the works. I'm not sure of the
priorities on the porting process, what's in 4.2 and what comes in 4.3
or later...

> However, if I do that, wouldn't it be true that I would have to run all
> my tests headless because, if I read things right, none of the tptp
> tools I've plugged into my Eclipse workbench at this point (tptp 4.1)
> can communicate with the new agent controller?

You're right that as of 4.1, the Eclipse workbench doesn't communicate
with the new AC. Very true. For 4.2 most of the communication is done by
the new AC implementing "backward compatibility"... the Eclipse
workbench just continues doing what it does, and the new AC has new
"transport layers" (CCTL, client compatibility TL and ACTL, agent
compatibility TL) that recognize that older protocol and interprets as
appropriate.

> So if I want to control my testing using the workbench, I have to write
> my agent using the old agent architecture. Is that correct?

Have to? No, I wouldn't go that far. One of my colleagues took the
statistical analysis front end portion that you get when you run the
PerfmonAgent from the workbench (Profiling and Logging... => Statistical
Analysis) and "ported it" to work with a "new tech AC" port of
PerfmonAgent I did. Neither effort was trivial, but neither was
particularly difficult either. Both portions were distributed with
4.1... the PerfmonAgent port I did in the Samples distributed in the new
tech AC SDK, and the workbench side in the PerfmonPluginforAC that was
available in 4.1 via the "Eclipse Feature Plug-in" link just below the
new tech AC portion of the download page.

> thanks,
> Rick

So back to your analysis... if it's a "do it now, do it quick, and be
done with it", you'll probably want to use the RAC... but if you've got
a longer view with maintenance in mind, perhaps you want to bear some
up-front costs for longer term maintainability.
--
RDS

Randy D. Smith randy (dot) d (dot) smith (at) intel (dot) com
Eclipse TPTP Committer, Platform Proj (data collection/agent controller)
Re: New Agent Questions [message #69622 is a reply to message #69597] Sun, 07 May 2006 12:33 Go to previous messageGo to next message
Eclipse UserFriend
Originally posted by: user.domain.invalid

Randy D. Smith wrote:
> So far so good. I agree with your analysis. The one glitch in your
> thinking is the HP/UX portion... so far the new AC is only ported to
> Intel architectures, though more are in the works. I'm not sure of the
> priorities on the porting process, what's in 4.2 and what comes in 4.3
> or later...
>
Thought I saw instructions for linux in a
how-to-compile-it-all-from-source read-me I found in CVS. If it runs
under linux, then getting it to hpux might be doable without a major
hair pulling session. Depends I guess on whether there are massive
amounts of #ifdef LINUX and #ifdef WINDOWS sections in the code. I've
tried porting open source stuff before and given up because of the
incomprehensible ifdef sections. My experience with Eclipse code has
been better than that however.

>> So if I want to control my testing using the workbench, I have to
>> write my agent using the old agent architecture. Is that correct?
>
>
> Have to? No, I wouldn't go that far. One of my colleagues took the
> statistical analysis front end portion that you get when you run the
> PerfmonAgent from the workbench (Profiling and Logging... => Statistical
> Analysis) and "ported it" to work with a "new tech AC" port of
> PerfmonAgent I did. Neither effort was trivial, but neither was
> particularly difficult either. Both portions were distributed with
> 4.1... the PerfmonAgent port I did in the Samples distributed in the new
> tech AC SDK, and the workbench side in the PerfmonPluginforAC that was
> available in 4.1 via the "Eclipse Feature Plug-in" link just below the
> new tech AC portion of the download page.
I'll look into those.

> So back to your analysis... if it's a "do it now, do it quick, and be
> done with it", you'll probably want to use the RAC... but if you've got
> a longer view with maintenance in mind, perhaps you want to bear some
> up-front costs for longer term maintainability.
It is a do it now-quick type effort, mainly because I'm trying to sell
this as a solution to a manager who is under a tight deadline and needs
his coders doing their unit testing in a uniform way. The main unix
geek on the team wants me to just go with the traditional unix script
test setup, but that would bore me to tears, so I'm trying to get this
up and going to relieve my own fears that the manager will pull the plug
on me.

Thanks for the pointers.
Rick
Re: New Agent Questions [message #69756 is a reply to message #69622] Mon, 08 May 2006 13:02 Go to previous messageGo to next message
Guru Nagarajan is currently offline Guru NagarajanFriend
Messages: 45
Registered: July 2009
Member
Rick,
Adding to what Randy mentioned - take a look at the plugin
org.eclipse.tptp.platform.jvmti.client for the access to the new Agent
controller and frameworks and the reuse of the TPTP UI components.

Thanks,
Guru
rkm wrote:
> Randy D. Smith wrote:
>> So far so good. I agree with your analysis. The one glitch in your
>> thinking is the HP/UX portion... so far the new AC is only ported to
>> Intel architectures, though more are in the works. I'm not sure of the
>> priorities on the porting process, what's in 4.2 and what comes in 4.3
>> or later...
>>
> Thought I saw instructions for linux in a
> how-to-compile-it-all-from-source read-me I found in CVS. If it runs
> under linux, then getting it to hpux might be doable without a major
> hair pulling session. Depends I guess on whether there are massive
> amounts of #ifdef LINUX and #ifdef WINDOWS sections in the code. I've
> tried porting open source stuff before and given up because of the
> incomprehensible ifdef sections. My experience with Eclipse code has
> been better than that however.
>
>>> So if I want to control my testing using the workbench, I have to
>>> write my agent using the old agent architecture. Is that correct?
>>
>>
>> Have to? No, I wouldn't go that far. One of my colleagues took the
>> statistical analysis front end portion that you get when you run the
>> PerfmonAgent from the workbench (Profiling and Logging... =>
>> Statistical Analysis) and "ported it" to work with a "new tech AC"
>> port of PerfmonAgent I did. Neither effort was trivial, but neither
>> was particularly difficult either. Both portions were distributed with
>> 4.1... the PerfmonAgent port I did in the Samples distributed in the
>> new tech AC SDK, and the workbench side in the PerfmonPluginforAC that
>> was available in 4.1 via the "Eclipse Feature Plug-in" link just below
>> the new tech AC portion of the download page.
> I'll look into those.
>
>> So back to your analysis... if it's a "do it now, do it quick, and be
>> done with it", you'll probably want to use the RAC... but if you've
>> got a longer view with maintenance in mind, perhaps you want to bear
>> some up-front costs for longer term maintainability.
> It is a do it now-quick type effort, mainly because I'm trying to sell
> this as a solution to a manager who is under a tight deadline and needs
> his coders doing their unit testing in a uniform way. The main unix
> geek on the team wants me to just go with the traditional unix script
> test setup, but that would bore me to tears, so I'm trying to get this
> up and going to relieve my own fears that the manager will pull the plug
> on me.
>
> Thanks for the pointers.
> Rick
Re: New Agent Questions [message #70179 is a reply to message #69622] Tue, 09 May 2006 15:13 Go to previous message
Randy D. Smith is currently offline Randy D. SmithFriend
Messages: 394
Registered: July 2009
Senior Member
rkm wrote:
> Randy D. Smith wrote:
>
>> So far so good. I agree with your analysis. The one glitch in your
>> thinking is the HP/UX portion... so far the new AC is only ported to
>> Intel architectures, though more are in the works. I'm not sure of the
>> priorities on the porting process, what's in 4.2 and what comes in 4.3
>> or later...
>>
> Thought I saw instructions for linux in a
> how-to-compile-it-all-from-source read-me I found in CVS. If it runs
> under linux, then getting it to hpux might be doable without a major
> hair pulling session. Depends I guess on whether there are massive
> amounts of #ifdef LINUX and #ifdef WINDOWS sections in the code. I've
> tried porting open source stuff before and given up because of the
> incomprehensible ifdef sections. My experience with Eclipse code has
> been better than that however.

Right, in our bin directory there is a readme.txt that we try to keep
up-to-date that contains build instructions.

You're also right that in general Eclipse code is more portable... but
then anything highly Java-centric should be.

The problem is that the AC and other *native* code is... well... native.
That's the primary reason for the inclusion of the C++ (and even C...
and occasionally even an asm call or two) code.

We took great pains to localize machine dependencies, and I think did a
pretty good job all things considered. However, there are machine
dependencies.

Then you get to the division of labor and the reality that it's not good
enough just to port it; we want to test and maintain it as well.
Currently HP/UX support is over on the IBM side of our team, the source
for the original RAC. The Intel portion of the team (that I'm a part of)
is responsible for the new tech AC, and we do the native builds for the
Intel architectures.

What we've found is that the porting is not difficult... it's just a
matter of allocating resources. Make particular note of the "call for
votes" for the 4.3 planning process for TPTP... if you have votes to
use, you might want to apply some to the HP/UX porting bugzilla
(presuming one exists... if it doesn't, then filing one would be the
best start!).

--
RDS

Randy D. Smith randy (dot) d (dot) smith (at) intel (dot) com
Eclipse TPTP Committer, Platform Proj (data collection/agent controller)
Previous Topic:[AGR] focus command generation
Next Topic:AGR: Can't record anything on a Spinner widget
Goto Forum:
  


Current Time: Mon Dec 22 02:58:29 GMT 2014

Powered by FUDForum. Page generated in 0.01795 seconds
.:: Contact :: Home ::.

Powered by: FUDforum 3.0.2.
Copyright ©2001-2010 FUDforum Bulletin Board Software