[orbit-user] trace_oml2 bottleneck?

[Max Ott]

That's quite likely as their is a design flaw in the current OML library if you collect every sample. Technically, it is not loosing any packets, but  the network pipe is blocking the collection - I know, the observed result is the same. 

Please do me a favor and run trace_oml2 with the --oml-file flag instead of --oml-server. --oml-fil will write all the measurements into a local file. That should be as fast as tcpdump. If that works, there is a work around using the OML Proxy server on each node (check with Guillaume).

Jolyon has already started to address this in an experimental branch,  but robust and parameterisable  load shedding within a very open framework as implemented in OML isn't easy - you'll find a lot of papers on this topic in the stream database community. 

For those wanting to know more on what happens under the hood in OML we just had a paper accepted for Tridentcom on this topic (check with Guillaume for the full paper).

Thanks for bringing this up. We love to hear from anyone who is running into issues. There is considerable and continuos engineering and development effort going into all the Orbit/OMF tools. So if someone tells you that something didn't work for them in the past, better check, it may have already been addressed. On the other hand, we have a long list of things we want to do, getting feedback from all you users on what is important for you makes it easier for us to prioritise the right things.

Cheers,
-max

On 30/01/2010, at 8:23 AM, Akash Baid wrote:

> Hi,
> 
> I am using the trace_oml2 application which is bundled with the OML client to log WiFi traces in the promiscuous mode. On an adjacent node, I am using tcpdump to essentially do the same thing. From a few experiments, the observation I have is when the bitrate of the sniffed data is let's say 1Mbps, both the methods capture a similar number of packets but when it increases to 54Mbps, the trace_oml2 loses a lot of packets (> 70%) while tcpdump seems to have the large number of packets received. 
> 
> I was wondering if this is a bottleneck while entering data into the OML database or something else that I might be observing? Would appreciate any help.
> 
> Thanks,
> Akash
> 
> 
> _______________________________________________
> orbit-user mailing list
> orbit-user at orbit-lab.org
> http://orbit-lab.org/cgi-bin/mailman/listinfo/orbit-user