[c-nsp] Effect of simultaneous TCP sessions on bandwidth

[Youssef Bengelloun-Zahr]

Le 11 nov. 2013 à 06:47, Brad Gould <bradley at internode.com.au> a écrit :

> Apply a shaper (not a policer) towards the service provider at each end of 95Mbps or so (will probably require tweaking).

Tried that, I applied a rate-limiter on my port facing my provider. No change...

> 
> Single TCP session is probably managing to balance itself into the ~100Mbps circuit.
> 
> Two/many TCP sessions are probably bursting into a policer (and effectively each other) often enough to ruin performance.

We tested with/without our provider policers on their equipments, no change.

When policers applies them, he confirms policers aren't being trigered by the flows we send.

> 
> Brad
> 
> On 10/11/2013 4:12 PM, Youssef Bengelloun-Zahr wrote:
>> Hello community,
>> 
>> Need your help and hands on experience to shed some light on some problem
>> I'm facing.
>> 
>> We have contracted a Layer 2 ethernet connection hand-off between our DC
>> (Frankfurt) and a customer site (Hamburg) with a carrier.
>> 
>> Carrier provides us with an ethernet MPLS pipe up to a DC in hamburg and
>> relies on a third party local loop provider to extend it up to customer
>> site. Nothing new under the sun here.
>> 
>> We have been testing this connection because we think we are facing
>> bandwidth issues. Let me summarize our results :
>> 
>>    - Carrier claims E2E Ethernet RFC2544 passed : we have been to check the
>> results and they seem OK,
>> 
>>    - UDP traffic reaches up to 95 Mbits/s for one way streams (both ways)
>> and simaltaneous bi-directionnal streams,
>> 
>>    - TCP traffic reaches up to 90 Mbits/s for one way streams (both ways),
>> 
>>    - TCP traffic hits some kind of limit and isn't able to achieve more
>> than 40-60 Mbits/s in average      <=== That's the problem we are facing.
>> 
>> One bit of information I think is relevant :
>> 
>>     - FRA Handoff between our provider and our PE is using a GigE port,
>> 
>>     - HBG Handoff between our provider and local-loop provider is using a
>> Fast ethernet ports between their facing equipments,
>> 
>>     - CE in Hamburg is a Fast Ethernet port and is forced with 100 Full
>> duplex,
>> 
>> 
>> We have carried tests with multiple devices directly connected behind our
>> PE in FRA and carrier's CE in HBG, results are always the same.
>> 
>> In the end, we connected servers directly in order to suppress any uneeded
>> equipments on the path, tests we're carried using iPerf and some other
>> tools.
>> 
>> We have been debugging this, no improvement. We have tried everything,
>> disabling all policiers, etc.... nothing nails it !
>> 
>> Our provider claims this is normal behavior for TCP. Does this sound normal
>> to you ?
>> 
>> Thanks for your help.
>> 
>> Best regards.
>> _______________________________________________
>> cisco-nsp mailing list  cisco-nsp at puck.nether.net
>> https://puck.nether.net/mailman/listinfo/cisco-nsp
>> archive at http://puck.nether.net/pipermail/cisco-nsp/
> 
> 
> -- 
> Brad Gould, Network Engineer
> iiNet / Internode
> P: +61 8 8228 2999
> bradley at internode.com.au
>