perfsonar development work

[Verena Venus <>]

Hello Sven,

Am Friday 05 January 2007 16:59 schrieb Jeff W. Boote:
> Stephan Kraft wrote:
> > Hi Sven,
> >
> > our office is closed during holidays, so we haven´t a chance to answer
> > your questions. Next week, we all will be back (or in Cambridge) to
> > solve the problems.
> >
> > Hope Verena will answer them.
>
> I can answer a couple of these questions...
>
> > Sven Ubik schrieb:
> >> Hi,
> >>
> >> tried bwctl installation at
> >> http://www.win-labor.dfn.de/bwctldemo/bwctl_demo.php
> >>
> >> and I have a few questions or notes:
> >>
> >> 1) What is difference between:
> >>
> >> "Buffer size (in byte) (Size of read/write buffer for udp)" and
> >> "Window size (in byte) (Size of tcp window / udp socket receive buffer)"
>
> Buffer size is the size of the 'user space' buffer that is read/written in
> the iperf application. This is for both TCP and UDP. (For UDP this also
> determines the packet size because it proscribes the full packet payload.)
>
> Window size is really the max send/recv socket buffer size as set by
> setsockopt and SO_RCVBUF and SO_SNDBUF. And, it is not actually used for
> UDP - but I believe iperf still takes the -w option without complaint.
>
> >> Window size does not have sense for UDP (which is mentioned on the
> >> second line). And you can never set window size for TCP, it is a dynamic
> >> property of TCP protocol. You can only set the upper limit of socket
> >> buffers, which then determine the upper limit of window size.
> >>
> >> On the other hand, you can also set the volume of data sent by the
> >> application at once, which can affect resulting throughput.
> >>
> >> Is the above mentioned "buffer size" what you set by sysctl or
> >> setsockopt() or the resulting real buffer size after internal
> >> adjustments in Linux kernel (which also depends whether it is 2.4 or
> >> 2.6)?
> >>
> >> What are default values?
>
> The default Buffer size in bwctl/iperf is:
>
> TCP: 8 KB
> UDP: 1470 bytes
>
> The default Window size is per-system tunable using sysctl. The actual used
> value that was used is in the iperf/bwctl output data. (But, I don't know
> if that is exposed by this web service.)
>
> Sorry, I can't help with the rest of the questions - they are specific to
> the DFN deployment. (VV - let me know if it looks like any of these
> problems are due to bwctl bugs. I would be happy to try and fix them. Along
> those lines, there is a bug in the current released version that makes the
> 'fuzzfactor' option not work correctly, so if these hosts are not properly
> synced with ntp that could be at least part of the problem.)
>
> jeff
>

Several of the boxes were not running properly, I fixed them all, and it is 
now working fine for me.

> >> 2) I tried first:
> >>
> >> Prague IPv4 -> Amsterdam IPv4 and I got:
> >>
> >> BWCTL Error: bwctl: Unable to connect to 192.87.102.210: Connection
> >> refused
> >>
> >> then I tried Paris -> Athens and I got:
> >>
> >> BWCTL Error: bwctl: Unable to connect to 194.177.210.70: Connection
> >> refused
> >>
> >> then I tried Paris -> Rome and I got:
> >>
> >> BWCTL Error: RECEIVER START RECEIVER END
> >>

This seems to be a BWCTL error, maybe due to the sync problem Jeff mentioned. 

> >> then I tried Sofia -> Zagreb and I got:
> >>
> >> BWCTL Error: bwctl: Unable to connect to 193.198.229.166: Connection
> >> refused
> >>
> >> then I tried Amsterdam ->  Athens for UDP (previous tests were for TCP)
> >> and I got:
> >>
> >> SOAP error:
> >> HTTP Error: Couldn't open socket connection to server
> >> http://192.87.102.210:8090/, Error (111): Connection refused
> >>
> >> then I tried Gent -> Ljubjjana for UDP and I got:
> >>
> >> BWCTL Error: RECEIVER START 3376807663.317963: iperf -B 193.2.63.2 -P 1
> >> -s -f b -m -p 5001 -u -t 10
> >>
> >> I did not find any combination that works.
> >>

As I have now fixed the boxes, it should work again. Please let me know, if 
you still see any problems.


> >> 3) When I do one measurement and go back to do another measurement, all
> >> items in the form are lost and I have to fill in everything again. Is it
> >> possible to remember the latest values?
> >>
> >> 4) When I find that throughput was low between some two points, do I get
> >> any feedback, that allows me to determine why it was low?
> >>

At the moment, we are not sure, if the development of the webclient as single 
tool is reasonable. If it gets integrated for example into perfSONAR UI, 
these features should be part of perfSONAR UI, or whatever other perfSONAR 
webclient it is to be integrated.

Thanks for your feedback! I hope all your questions are answered, feel free 
to 
ask further, if you want to know more.

Regards,
Verena
 
> >> Regards,
> >>
> >> Sven

-- 
Verena Venus               Tel: +49 9131 852-8705
DFN Labor

Regionales Rechenzentrum Erlangen
Martensstrasse 1
91058 Erlangen