Cisco Support Community
Showing results for 
Search instead for 
Did you mean: 

Welcome to Cisco Support Community. We would love to have your feedback.

For an introduction to the new site, click here. And see here for current known issues.

New Member

TCP Window size


We have recently had a 50Mb ethernet service provisioned between 2 POPS - Latency is approx. 13m/sec.

Testing the service, we are only able to achieve ~800K/sec (FTP+HTTP between 2 Linux boxes) in a single session(If we run multiple tests simultaneously, each single session is able to achieve approx. 500K/sec, but the aggregate total gets up to ~20Mb(If we ran more sessions, I can only assume we would be able to reach 50Mb)

We have an existing ATM service(13Mb), and performing the same test between the 2 Linux boxes, we are able to achieve 1.1->1.2 Mb/sec.

The Ethernet provider has tested the service(RFC 2544) and confirmed 50Mb throughput.

Ethernet provider is suggesting we modify tcp window size - I can see this can be done in global config mode (ip tcp window-size 80000), but this would affect all Interfaces? Would there be any adverse affects from doing this?

Ethernet provider has also suggested we implement rate-limiting on our Interfaces...both ends are connected to 2950's, which are then trunked up to 7204VXR's as dot1q Ints....I'm assuming we will need to do the rate-limiting on the 7200's?

Thanks in advance.

New Member

Re: TCP Window size


To test the bandwidth , I think u shud use udp based tools, its avalable on Internet. Change the tcp window size shud not affect nromally, but it also depends upon application which you run. It shud be same on both ends. Better test this with application you need to run..




Re: TCP Window size


Adjusting TCP window size on the router will only affect router originated packets. Routers do not normally care about window size settings of through connections.

It is the Linux box window size you need to look at. Newer Linux versions support window size scaling which allows for larger than 65536 byte windows sizes. Your 50mb pipe with 13ms latency will take 162,500 bytes (BPD bandwidth delay product) to fill, so a single connection would need to use window sizes at least that large to attain 50 mb/s throughput.

Since your tested throughput is much less than the default window sizes would handle I would suspect lost packets and re-transmits. Do you have a network analyzer (Sniffer or Wireshark) you can use? You need to determine if the packet are being dropped in your network. Your rates are so low that a duplex mismatch could be the culprit. Check your switch ports for errors.

Another possibility for testing thoughput on the circuit is TTCP between the routers. This is a hidden utility in some versions of IOS that can be used to test throughput. You set one router as the transmitter and the other as receiver. Enter TTCP from the enable prompt to see if you have it:


transmit or receive [receive]:

Please rate helpful posts.


CreatePlease to create content