slow speeds over MPLS link

[carl_townshend]

Hi All

We have a 15Mbit MPLS link between the UK and USA, The latency on this is around 110ms

We seem to be unable to get more than 6-7Mbit out the link and no more.

Why would this be? what is the theoretical maximum speed with that latency?

Cheers

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Theoretical maximum speed should be the 15 Mbps.

A common cause for being unable to push to full rate, many older TCP implementation default RWIN isn't sized sufficiently for a LFN (long fat network).

[carl_townshend]

what about the Bandwidth delay product? with a default Win size of 65535 etc it says I can only get 4.5 meg a second, does this take into account scaling factor ?

[pwwiddicombe]

On a single connection, the latency will cause you the apparent ceiling you have discovered.  Now, you might find you can run multiple transfers at that rate (from what you are seeing, that looks like 2 concurrent "transfers" could run at the 7 Meg rate.   With multiple transfers going on concurrently, check the interface counters on the router that connects to the MPLS link, and you might find it's close to 15 Meg, if the MPLS provider has all settings in order. 

If NOT, then verify interface counters for CRC's or overruns and you might have duplex mismatches.

This is the difference between "speed" and "bandwidth".  With no latency, they're pretty much the same. 

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

RWIN is what is used to support BDP.

65K would support various bandwidths, higher bandwidth with less delay, lower bandwidth with more delay.

Scaling, if both sides support and enable it, support a huge RWIN, which can support very large BDPs.  65K, with scaling, would depend on the scaling factor to indicate the actual RWIN.  For example 2x with 65K is 128K, 4x is 256K, etc.

BTW, although scaling supports huge BDP, you still need to account for slow start, i.e. a small transaction flow might never scale up to maximum transmission rate before the data is transferred.  You also need to account for the impact of last packets, where even one lost packet causes a very slow transmission acceleration rate because of congestion advoidance.  Again, a single TCP flow might complete before it gets up to the theoretical rate.

Unless you get into experimental transmission protocols, often the easiest way to drive a LFN is multiple concurrent TCP flows.

[carl_townshend]

what would be defined as a long fat network?

Also with the speed and latency I mentioned, is the speed im getting about right?

if scaling was used could 1 connection be able to fill this line ?

cheers

[Joseph W. Doherty · Accepted solution]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Carl, have you tried searching the Internet for answers to your last posting's questions?

For example, this Wiki article (https://en.wikipedia.org/wiki/Bandwidth-delay_product) says RFC 1072 defines LFNs.

There are (web) tools that will calculate BDP.  For example:

http://www.speedguide.net/bdp.php

https://www.switch.ch/network/tools/tcp_throughput/