[j-nsp] Antwort: Re: Antwort: Re: Antwort: Re: Bad TCP Session Performance

[Hannes Gredler]

some node must be dropping ... you need to figure out hwo is dropping:
can you tcpdump the FTP session and check with tcptrace *why*
it is that slow ? - you should see a throughput of 140MBit/s.

/hannes

jens.hoffmann at ddkom.de wrote:
> I'm using various applications for the test :
> 
> - end to end ftp-transfer
> - ftp-transfer between two Juniper Nodes via STM-1 or aggregated SONET
> 
> The following transfer shows the bandwidth limitation :
> 
> ftp> get /var/tmp/jinstall-7.2R1.7-domestic-signed.tgz
> local: /var/tmp/jinstall-7.2R1.7-domestic-signed.tgz remote:
> /var/tmp/jinstall-7.2R1.7-domestic-signed.tgz
> 200 PORT command successful.
> 150 Opening BINARY mode data connection for '/var/tmp/jinstall-7.2
> R1.7-domestic-signed.tgz' (66618900 bytes).
> 100%
> |******************************************************************************************|
>  65057 KB    00:00 ETA
> 226 Transfer complete.
> 66618900 bytes received in 29.96 seconds (2.12 MB/s)
> 
> 
> 
> The RTT between these nodes is about 3ms.
> The STM-1-Link has a load of about 28-30 %.
> 
> Jens
> 
> Hannes Gredler <hannes at juniper.net> schrieb am 03.04.2006 10:08:01:
> 
> 
>>what kind of test-application do you use ?
>>
>>
>>jens.hoffmann at ddkom.de wrote:
>>
>>>The test runs as follows :
>>>
>>>Host-1 === fe === M7i ===aggregated SONET 2xSTM-1 === M5 === fe(802.1Q)
> 
> ===
> 
>>>CATALYST3550 === Host-2
>>>
>>>or
>>>
>>>Host-1 === fe === M7i ===aggregated SONET  2xSTM-1 === M5 === fe ===
> 
> Host-2
> 
>>>or
>>>
>>>Host-1 === fe === M10i ===SONET 1xSTM-1=== M5 === fe === Host-2
>>>
>>>Kind Regards
>>>Jens
>>>
>>>Hannes Gredler <hannes at juniper.net> schrieb am 03.04.2006 09:51:35:
>>>
>>>
>>>
>>>>ok 7.0R2.7 is not affected by the TCP problem
>>>>that was mentioned by luiz;
>>>>
>>>>did you do perform your throughput test from the RE to some host on the
>>>
>>>net ?
>>>
>>>
>>>>/hannes
>>>>
>>>>jens.hoffmann at ddkom.de wrote:
>>>>
>>>>
>>>>>Hi Hannes,
>>>>>
>>>>>I'm using the same JUNOS 7.0R2.7 on each device (M7i, M10i, M5 and
>>>
>>>M10).
>>>
>>>
>>>>>The end to end RTT is not conspicuous.
>>>>>
>>>>>Model: m7i
>>>>>JUNOS Base OS boot [7.0R2.7]
>>>>>JUNOS Base OS Software Suite [7.0R2.7]
>>>>>JUNOS Kernel Software Suite [7.0R2.7]
>>>>>JUNOS Packet Forwarding Engine Support (M7i/M10i) [7.0R2.7]
>>>>>JUNOS Routing Software Suite [7.0R2.7]
>>>>>JUNOS Online Documentation [7.0R2.7]
>>>>>JUNOS Crypto Software Suite [7.0R2.7]
>>>>>
>>>>>Kind Regards
>>>>>Jens
>>>>>
>>>>>
>>>>>Hannes Gredler <hannes at juniper.net> schrieb am 03.04.2006 09:20:27:
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>jens,
>>>>>>
>>>>>>what JUNOS S/W do you run ?
>>>>>>
>>>>>>/hannes
>>>>>>
>>>>>>jens.hoffmann at ddkom.de wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>>>Hi,
>>>>>>>
>>>>>>>I have a litlle bit tricky problem.
>>>>>>>
>>>>>>>Without packet loss, large latency and lack of WAN bandwidth  I'm
>>>>>
>>>>>recognize
>>>>>
>>>>>
>>>>>
>>>>>>>a limited TCP session bandwidth at about 16-20 Mbps for every single
>>>>>
>>>>>TCP
>>>>>
>>>>>
>>>>>
>>>>>>>session.
>>>>>>>
>>>>>>>At first I had the suspect that the problem is caused by enabled
> 
> flow
> 
>>>>>>>control on some ethernet interfaces. But the problem appears also
> 
> with
> 
>>>>>>>disabled flow control.
>>>>>>>Policers are also not configured on the path of the session.
>>>>>>>
>>>>>>>I'ld be very pleased for some helpful advice.
>>>>>>>
>>>>>>>Kind Regards,
>>>>>>>Jens