perfsonar development work

[Loukik Kudarimoti <>]

Roman Lapacz wrote:
> Vedrin Jeliazkov wrote:
> > Hi,
> >   
>
> Hi,
>
> > I've run some simple tests against the upgraded RRD MA at ESnet
>
> Do they use eXist to store the metadata configuration file?
>
> >  in an attempt
> > to compare its performance against the older release. Here are the 
> > timings
> > I've got:
> >
> > 1. One MetadataKeyRequest (summary) [sec]   
> > 2. Number of Interfaces (NoI)   
> > 3. (NoI x 2) SetupDataRequests (detailed summary) [sec]
> > 4. Average SetupDataRequest [msec]
> > 5. Average Link latency (RTT) [msec]
> > 6. Implementation
> >     
> >         (1)    (2)    (3)    (4)    (5)    (6)   
> >     
> > ESNET        180    415     468     564    202    Java 0.1 (EGEE tag)
> > ESNET        296    419    1600    1909    206    perfSONAR-1.0
> >  
Wow. I was expecting quite the opposite. The EGEE tag creates a lot of 
tomcat logs because of system.out (close to 1 gig a day) and I thought 
this would have severely decreased its performance. perfSONAR 1.0 on the 
other hand doesn't create so many logs I believe. I can imagine how good 
(comparitively) EGEE tag's results would be if we had removed all those 
unnecessary system.out logs. Hence, I believe the latest snapshot should 
perform better than EGEE tag (and not just remain similar)
> Unfortunately the release 1.0 does not contain recent changes, which I 
> hope, improve the performance. It would be interesting to see the 
> tests of RRD MA from svn trunk or the next release (question to the 
> release team: is there a timetable for this?).
I can set this up soon for GEANT2 but the number of interfaces for 
GEANT2 aren't that many.

Joe, since you probably have the highest number of interface 
configurations, would it be possible for you to install the recent 
snapshot as well? Of course, once the tests are done, the service can be 
uninstalled. Another option could be to mail me your config file and I 
can use it for test purposes (I can modify all the keys to point to the 
same dummy rrd file).

I believe the configuration file can remain more or less the same (I 
think the key needs one more parameter which specifies the data units 
used by the service to provide data). There are very minute schema 
improvements in the latest snapshot. (timeUnits and valueUnits 
attributes present for each datum) that perfSONAR UI or Vedrin's 
performance testing scripts would need to take care of.

Regards,
Loukik.
> BTW. Vedrin, are you a member of the release team? (I remember the 
> discussion in Cambridge with the decision to ask you to have you there 
> as you are a very active perfSONAR member).
>
>
> Roman
>
>
>
> > As you can see, the differences in the number of interfaces and the 
> > network
> > latency are insignificant, while MetadataKeyRequest's and 
> > SetupDataREquest's
> > processing time is much longer in the new service version, when 
> > compared to
> > the older one (multiplied by 1.6 and 3.4 respectively).
> >
> > If the service is running on the same hardware platform, than we could
> > conclude that there is a significant performance loss in the new 
> > version,
> > which is most probably due to some newly introduced scalability 
> > issue. Perhaps
> > some more comprehensive tests should be run in order to understand 
> > and resolve
> > this issue.
> >
> > Kind regards,
> > Vedrin
> >