Re: TCP Window Optiimization with LFN.

Small or unscaling TCP window sizes will greatly decrease your ability
to move data on LFNs. Here is a good link I use for calculating
expected throughput based on various parameters:

http://www.switch.ch/network/tools/tcp_throughput/index.html

The endpoints setting up the TCP session should have window scaling
enabled and have a large maximum window size. Some devices, even
fairly recent server OSes, seem to really want to use nothing larger
than 64 KB, which will give you really horrible throughput on longer
links. Grab a packet capture of a session and look at the TCP setup to
see if window scaling is enabled and what the scale factor is.
Wireshark will compute the actual window size for you.

John

On Thu, Aug 16, 2012 at 2:27 PM, Joseph L. Brunner
<joe_at_affirmedsystems.com> wrote:
> We tried that extensively in windows 2003 and other things - never worked or made a difference....
>
>
>
> -----Original Message-----
> From: nobody_at_groupstudy.com [mailto:nobody_at_groupstudy.com] On Behalf Of Dan Shechter
> Sent: Thursday, August 16, 2012 4:12 PM
> To: JB Poplawski
> Cc: ccielab_at_groupstudy.com
> Subject: Re: TCP Window Optiimization with LFN.
>
> It depends on what you are looking for: High BW, Latency issues or both.
>
>
> For a large window size,
> http://en.wikipedia.org/wiki/TCP_window_scale_option should do it. As long as there are no CRC errors on the path, there is no much harm enabling it by default.
>
> For latency issues of chatty protocols, we will all have to wait for the results of CERN experiments to see if we can send packets faster then the speed of light... ;) . For some applications, WAAS "reducing" the latency is much more important than BW.
>
>
> HTH,
> Dan #13685 (RS/Sec/SP)
> The CCIE troubleshooting blog: http://dans-net.com Bring order to your Private VLAN network: http://marathon-networks.com
>
>
>
>
>
> On Thu, Aug 16, 2012 at 9:07 PM, JB Poplawski <jb.poplawski_at_gmail.com>wrote:
>
>> Good afternoon,
>>
>> My token question out to the group as I await TAC.
>>
>> I have a network with Gb links between sites. Connected like such:
>> Site A < 7ms > Site B < 71ms > Site C <9ms> Site D
>>
>> End to end latency between Site A and Site D is 69ms. My numbers
>> above are the average given from CLI.
>>
>> I've been reading on LFNs (Long Fat Networks) and basically I need to
>> optimize the TCP Window size (calculated to 8625KB).
>>
>> My question, where do I implement it?
>> Site A and Site D? Or Site B and Site C?
>>
>> If I implement on Site B and Site C, how does that impact Site C to
>> Site D, etc?
>>
>> Any ideas?
>>
>> We previously had MPLS connection with WAAS which optimized traffic.
>> Now since we're running Gb, we removed WAAS (our WAAS boxes were only
>> rated to 50Mbps).
>>
>> Any suggestions? Am I on the right path?
>>
>> Any help is greatly appreciated.
>> JB
>>
>>
>> Blogs and organic groups at http://www.ccie.net
>>
>> ______________________________________________________________________
>> _ Subscription information may be found at:
>> http://www.groupstudy.com/list/CCIELab.html
>
>
> Blogs and organic groups at http://www.ccie.net
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html
>
>
> Blogs and organic groups at http://www.ccie.net
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html

Blogs and organic groups at http://www.ccie.net
Received on Thu Aug 16 2012 - 15:39:32 ART