[pacnog] Allot Netenforcer AC-401

["Alo Anesi" <Alo.Anesi@bluesky.as>]

All-

Thanks for the replies to my TCP/IP question.

On a related note, is anyone else using the Netenforcer from Allot Communications?

So far, we've had good experiences, but are now running into a problem bandwidth will start to degrade by about 2M. Restarting the flows clears the problem and bandwidth jumps back up to 6.5M and above, but we're having to do this daily. Support has been unable to find any problems, and I was hoping someone had any insight.

Thanks,
Alo	

-----Original Message-----
From: Joel Jaeggli [mailto:joelja@uoregon.edu] 
Sent: Sunday, November 19, 2006 6:57 AM
To: Alo Anesi
Cc: pacnog@pacnog.org
Subject: Re: [pacnog] TCP Send/Receive Window



Alo Anesi wrote:
> I have a 7.5Mb down by 2.5Mb up satellite connection with about 700ms latency. Would there be any reason to look into modifying the TCP Send/Receive window on my 7206VXR gateway router? Would I ever need to change it for this router? If so, how much bandwidth can I scale up to on a satellite connection?

If you need to move data from your router over the link then playing
with window scaling might be a good idea. However in this case the knob
you really want to tweak is actually on the end hosts since the router
isn't meddling with the receive window of flows going through it (there
are middle boxes that do this).

You probably already know this, but bandwidth delay product (the size
you need the window set to in order to achieve a given throughput given
a certain amount of latency is expressed as:

bandwidth × delay product = link capacity (which is also the size of the
receive window required to fill the link.

7.5Mb/s with 700 ms rtt = requires an rwin of about 640KB.

Of course you probably don't want to fill the link with only one flow,
but tcp slow-start will prevent that in the presence of other traffic
anyway. On some platforms (ie linux) you can set your tcp congestion
avoidance algorithm via a sysctl which in conjunction with a a larger
rwin can dramatically increase your link utilization for a given flow
but not all algorithm's play nice with others...

> Thanks,
> Alo
> 
> _______________________________________________
> pacnog mailing list
> pacnog@pacnog.org
> http://mailman.apnic.net/mailman/listinfo/pacnog

-- 
------------------------------------------------------------------------
Joel Jaeggli             Unix Consulting              joelja@uoregon.edu
GPG Key Fingerprint:   5C6E 0104 BAF0 40B0 5BD3 C38B F000 35AB B67F 56B2