[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: iperf / ftp / http TCP poor performance in one direction (UDP good)



It's probably not the memory then, but you can make sure by running
'free -m' and seeing if the system is too loaded.

>starts at win=6912 and grows to 165504 (This is where I start to see a heap of TCP Dup ACK and TCP segment of a reassembled PDU)), then increases to 353408 (Again, more TCP Dup ACK's), and looks to max out at 669056.
The max window you gave divided by the delay (67ms) is about 10
mega*byte* / second. This Is the speed you were experiencing, right?
Everything points to the window, then.

Did you check if A acknowledges every received segment? It could be
that it is a slower host and can't process all that data that fast.

Guido

On Fri, Apr 12, 2013 at 1:03 AM, John Elliot <johnelliot67@hotmail.com> wrote:
> Thanks for the reply.
>
> POPB MEM - 2Gb
> POPA MEM - 4Gb
>
>
> Wireshark: (During a wget)
>
> #tcp window size in the "bad" direction:
>
> starts at win=6912 and grows to 165504 (This is where I start to see a heap
> of TCP Dup ACK and TCP segment of a reassembled PDU)), then increases to
> 353408 (Again, more TCP Dup ACK's), and looks to max out at 669056.
>
> #tcp window size in the "good" direction:
>
> starts at win=6912 and looks to max out at 1995392 and has very few TCP Dup
> ACK and TCP segment of a reassembled PDU
>
>
>>
>> Hi
>>
>> >Provider of the link is stating it is a tcp window issue
>> I think that would be a bit weird, TCP is a very robust protocol
>> capable to adapt pretty much any bandwidth / latency situations.
>> What are the specifications on the hosts? Maybe POPA has too little
>> RAM and can't allocate a big enough TCP window?
>> Can you use wireshark or some other tool to capture the TCP segments
>> and look at the advertised window?
>>
>> Greetings, Guido.
>>
>> On Fri, Apr 12, 2013 at 12:37 AM, John Elliot <johnelliot67@hotmail.com>
>> wrote:
>> > Hi Guys,
>> >
>> >
>> > I have a 40Mb connection between 2 POPs, and 2 Debian servers connected
>> > at
>> > each end(L2, same broadcast domain) - I am seeing consistent poor
>> > performance with TCP in only one direction.
>> >
>> >
>> > Latency between the POP's is ~67m/sec
>> >
>> >
>> > POPA->POPB UDP(IPERF) - consistent 38.5Mb/sec
>> >
>> > POPA->POPB TCP(IPERF) - fluctuates ~30Mb/sec -> 32Mb/sec
>> >
>> > POPA->POPB TCP(FTP) - fluctuates ~30Mb/sec -> 32Mb/sec
>> >
>> > POPA->POPB TCP(WGET) - fluctuates ~30Mb/sec -> 32Mb/sec
>> >
>> > POPB->POPA UDP(IPERF) - consistent 38.5Mb/sec
>> >
>> > POPB->POPA TCP(IPERF) - fluctuates ~10Mb/sec -> 15Mb/sec
>> >
>> > POPB->POPA TCP(FTP) - fluctuates ~10Mb/sec -> 15Mb/sec
>> >
>> > POPB->POPA TCP(WGET) - fluctuates ~10Mb/sec -> 15Mb/sec
>> >
>> >
>> > Both Deb boxes are running the same kernel (2.6.32-5), and I have tried
>> > adjusting tcp window without seeing any improvements...all ports are
>> > running
>> > full duplex, and Im not seeing any errors/drops, nor packet-loss on the
>> > link.
>> >
>> >
>> > Any suggestions are greatly appreciated (Provider of the link is stating
>> > it
>> > is a tcp window issue) -If that is the case, if someone can provide some
>> > suggested tweaks/changes to the Deb servers, It would be fantastic!
>> >
>> > Cheers.
>> >
>> >
>> >
>> >


Reply to: