Skip to content

XAware Community

Get the Flash Player to see this player.
Flash Image Rotator Module by Joomlashack.
XAware 5.6
Need Help
Webinars and Events
Advanced Tutorials
Webinars and Events

You are here: Home arrow Forums
XAware Community Forums
Welcome, Guest
Please Login or Register.    Lost Password?
Re:Load stress testing & performance tuning (1 viewing) (1) Guest
Go to bottom Post Reply Favoured: 0
TOPIC: Re:Load stress testing & performance tuning
#5293
kvandersluis (Admin)
Admin
Posts: 402
User Online Now Click here to see the profile of this user
Re:Load stress testing & performance tuning 11 Years ago Karma: 13  
Hi Noam,

Paul made the comparisons in our environment, and let me know earlier this week that our results are fairly consistent with yours, for your specific use case. It does appear that what I mentioned earlier is true, that you do suffer a penalty based on the interpretive nature of the engine. The configuration you are using does not have any major issues, so your numbers are likely what you can expect using the current release. We found that you can improve speed a small amount (a couple of percentage points) by using pass through mode on your queries (set xa:oper="pass_through" on the xa:request element). This avoids one network request in the prepare/execute cycle used by the engine's default prepared statement mode.

We are currently looking at a broad range of performance improvement possibilities, ranging from specific SQL optimizations that match your use case (fairly simple queries, simple XML results), to broad-based improvements like code generation, providing an option to compile BizDocuments down to code. Some improvements will be part of the 6.0 release, others will have to be scheduled further out.

It would be good to hear from you the criticality of your performance metrics. If you can give us a specific target performance level that would lead you to choose XAware for your project, then this will help drive resource allocation decisions for which improvements we focus on.

Thanks,

-Kirstan
 
Report to moderator   Logged Logged  
  The administrator has disabled public write access.
#5295
n_shien (User)
XAware user
Posts: 15
User Offline Click here to see the profile of this user
Re:Load stress testing & performance tuning 11 Years ago Karma: 0  
Hi Kirsten an Paul,

At a guess, we are looking for an improvement from 3.45 to 0.2 - 0.1 seconds.

These Data Services usually form a part of composite services, 99% of our service are currently synchronous so any increase in responses of the composed services can reduce the overall performance of calling composite service.

As you can see we have been investigating XAware with the intention to progress from POC to Production, and we see the potential of XAware with regards to time to market, though we need to improve the response time.

Just as a thought would it be possible to use a java callout to a Spring JDBC template instead of calling the xdr, would that improve performance?

Regards,

Noam.
 
Report to moderator   Logged Logged  
  The administrator has disabled public write access.
#5297
kvandersluis (Admin)
Admin
Posts: 402
User Online Now Click here to see the profile of this user
Re:Load stress testing & performance tuning 11 Years ago Karma: 13  
Hi Noam,

I think there must be a confusion in the numbers somewhere. Maybe I'm not understanding your numbers. Paul's testing showed his custom-coded solution was running roughly 3 times faster for this use case. This is consistent with your report showing throughput and transaction times on your end. I'm not clear on where the 3.45 comes into play? You don't mean 3.45 seconds per XAware transaction, do you? That would be way too high, and would indicate a configuration issue somewhere. Your report shows about 42 XAware transactions per second, which should equate to about .03 seconds per transaction.

If you are indeed seeing 3.45 seconds, the most likely culprits to look for are 1) a configuration that initializes the engine on every invocation, 2) logging that is set too high (DEBUG level pays a very high penalty and is never appropriate for production... INFO level is recommended).

Also, I see from your report that the XAware test used 1 agent, and the coded solution used 2 agents. Is there some reason for that? If these are client threads making requests, it seems you would want the agent count to be the same for a comparison.

Thanks,

-Kirstan
 
Report to moderator   Logged Logged  
  The administrator has disabled public write access.
#5304
n_shien (User)
XAware user
Posts: 15
User Offline Click here to see the profile of this user
Re:Load stress testing & performance tuning 10 Years, 12 Months ago Karma: 0  
Hi Kirstan,
I do appreciate the point about the agents and I will investigate how would this effect the results with our test team.

As for the 3.45 seconds is the average response time per second (green graph) that is what our test team measure for test exit for our operational acceptance.

I hope that this explains the 3.45.

Both applications have been deployed to Weblogic configured the same including connection pools etc, I'll look at logging levels though I would be surprised that this effect would be so substantial.

I'll get back to you regarding changing the agent number and the results from that.

Regards,

Noam.
 
Report to moderator   Logged Logged  
  The administrator has disabled public write access.
#5312
kvandersluis (Admin)
Admin
Posts: 402
User Online Now Click here to see the profile of this user
Re:Load stress testing & performance tuning 10 Years, 12 Months ago Karma: 13  
Hi Noam,

If I understand the report properly, the 3.45 second response time is for pages returned at the application level. Is this true? Its hard to guage where that time is going, since the XAware transactions seem to take less that 0.03 seconds. I'm guessing there are multiple calls per page, is that right?

As a quick performance test, I fired up SoapUI and was able to get 60 tranasactions per second over HTTP on one of the XAware examples, with a simple, non-indexed database query, and simple XML results.
 
Report to moderator   Logged Logged  
  The administrator has disabled public write access.
#5313
n_shien (User)
XAware user
Posts: 15
User Offline Click here to see the profile of this user
Re:Load stress testing & performance tuning 10 Years, 12 Months ago Karma: 0  
Hi Kirstan,
Our test team use a tool called silk performer, it basically acts the same as soapUI, except that it uses multiple IP for the requests.

We only call the services using SOAP over http.
We have used the same database for both tests with the same inputs for each test.

I am trying to implement a java callout using Spring JdbcDaoSupport and JdbcTemplate, I'll let you know what the resaults are when I have managed to get it running and tested.

Regards,

Noam.
 
Report to moderator   Logged Logged  
  The administrator has disabled public write access.
Go to top Post Reply
Powered by FireBoardget the latest posts directly to your desktop

Community Login

Get the Flash Player to see this player.
Flash Image Rotator Module by Joomlashack.
Commercial
Free Training
QuickStart Packages
Image 4 Title
Image 5 Title

Visit XAware.com