[gentoo-user] TCP Advanced Congestion Control -- any difference?

[gentoo-user] TCP Advanced Congestion Control -- any difference?

[Pandu Poluan]

Like the subject said: I am wondering if using a non-default TCP
Advanced Congestion Control makes any difference.

(The default is "cubic", but there are alternatives such as "htcp",
"hybla", and "yeah")

Any experiences?

Rgds,
--
FdS Pandu E Poluan
~ IT Optimizer ~

 • LOPSA Member #15248
 • Blog : http://pepoluan.tumblr.com
 • Linked-In : http://id.linkedin.com/in/pepoluan

Re: [gentoo-user] TCP Advanced Congestion Control -- any difference?

[Florian Philipp]

[-- Attachment #1: Type: text/plain, Size: 881 bytes --]

Am 20.10.2011 15:10, schrieb Pandu Poluan:
> Like the subject said: I am wondering if using a non-default TCP
> Advanced Congestion Control makes any difference.
> 
> (The default is "cubic", but there are alternatives such as "htcp",
> "hybla", and "yeah")
> 
> Any experiences?
> 

I tested it on the only situation I had where it was even remotely worth
the effort to try it: NFS over TCP via an old and overutilized router:
No measurable effect. I guess a web or mail server (read: something that
is not primarily bandwidth constrained and where latency matters) might
benefit more. But then again, how do you measure that reliably?

You also have to consider where the client might be. A long distance,
high bandwidth connection will benefit from different congestion control
mechanisms than a local low bandwidth connection.

Regards,
Florian Philipp


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: [gentoo-user] TCP Advanced Congestion Control -- any difference?

[Florian Philipp]

[-- Attachment #1: Type: text/plain, Size: 1071 bytes --]

Am 20.10.2011 18:11, schrieb Florian Philipp:
> Am 20.10.2011 15:10, schrieb Pandu Poluan:
>> Like the subject said: I am wondering if using a non-default TCP
>> Advanced Congestion Control makes any difference.
>>
>> (The default is "cubic", but there are alternatives such as "htcp",
>> "hybla", and "yeah")
>>
>> Any experiences?
>>
> 
> I tested it on the only situation I had where it was even remotely worth
> the effort to try it: NFS over TCP via an old and overutilized router:
> No measurable effect. I guess a web or mail server (read: something that
> is not primarily bandwidth constrained and where latency matters) might
> benefit more. But then again, how do you measure that reliably?
> 
> You also have to consider where the client might be. A long distance,
> high bandwidth connection will benefit from different congestion control
> mechanisms than a local low bandwidth connection.
> 
> Regards,
> Florian Philipp
> 

This paper and its references could be interesting.
http://research.google.com/pubs/archive/37486.pdf


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: [gentoo-user] TCP Advanced Congestion Control -- any difference?

[Pandu Poluan]

[-- Attachment #1: Type: text/plain, Size: 1253 bytes --]

On Oct 23, 2011 7:07 PM, "Florian Philipp" <lists@binarywings.net> wrote:
>
> Am 20.10.2011 18:11, schrieb Florian Philipp:
> > Am 20.10.2011 15:10, schrieb Pandu Poluan:
> >> Like the subject said: I am wondering if using a non-default TCP
> >> Advanced Congestion Control makes any difference.
> >>
> >> (The default is "cubic", but there are alternatives such as "htcp",
> >> "hybla", and "yeah")
> >>
> >> Any experiences?
> >>
> >
> > I tested it on the only situation I had where it was even remotely worth
> > the effort to try it: NFS over TCP via an old and overutilized router:
> > No measurable effect. I guess a web or mail server (read: something that
> > is not primarily bandwidth constrained and where latency matters) might
> > benefit more. But then again, how do you measure that reliably?
> >
> > You also have to consider where the client might be. A long distance,
> > high bandwidth connection will benefit from different congestion control
> > mechanisms than a local low bandwidth connection.
> >
> > Regards,
> > Florian Philipp
> >
>
> This paper and its references could be interesting.
> http://research.google.com/pubs/archive/37486.pdf
>

Thanks! I'll sure to study them tomorrow. Productive time at the office ;-)

Rgds,

[-- Attachment #2: Type: text/html, Size: 1746 bytes --]