TCP rate/congestion control
I noticed that Windows 10 now uses CTCP as the default TCP congestion/rate control algorithm, but Linux still defaults to the old Cubic algorithm. CTCP doesn't appear to be available on Ubuntu LTS at the moment, but there's a whole host of others to choose from. Has anyone here worked out which is the best one to use on typical consumer internet links in Australia? (ie. ADSL with 4-12 mbps and only minor packet loss) Available options on Ubuntu Xenial: - bic - cubic - cdg - dctcp - diag - highspeed - htcp - hybla - illinois - lp - probe - reno - scalable - vegas - veno - westwood - yeah
On 16/09/16 11:02, Toby Corkindale via luv-main wrote:
I noticed that Windows 10 now uses CTCP as the default TCP congestion/rate control algorithm, but Linux still defaults to the old Cubic algorithm.
CTCP doesn't appear to be available on Ubuntu LTS at the moment, but there's a whole host of others to choose from. Has anyone here worked out which is the best one to use on typical consumer internet links in Australia?
Over and above the rate control algorithm Linux has a bunch of features that make it work much better than a to-the-spec cubic implementation (not surprising with a bunch of large content providers like $EMPLOYER submitting their fixes upstream). Things like TCP pacing and the work from the bufferbloat folk have really improved things. https://fasterdata.es.net/host-tuning/linux/fair-queuing-scheduler/
On 16/09/16 18:33, Julien Goodwin via luv-main wrote:
On 16/09/16 11:02, Toby Corkindale via luv-main wrote:
I noticed that Windows 10 now uses CTCP as the default TCP congestion/rate control algorithm, but Linux still defaults to the old Cubic algorithm.
CTCP doesn't appear to be available on Ubuntu LTS at the moment, but there's a whole host of others to choose from. Has anyone here worked out which is the best one to use on typical consumer internet links in Australia?
Over and above the rate control algorithm Linux has a bunch of features that make it work much better than a to-the-spec cubic implementation (not surprising with a bunch of large content providers like $EMPLOYER submitting their fixes upstream).
Things like TCP pacing and the work from the bufferbloat folk have really improved things.
https://fasterdata.es.net/host-tuning/linux/fair-queuing-scheduler/
...and something that finally went public overnight is a new TCP rate control algorithm from some of the folk at $EMPLOYER, I for one am very much looking forward to this hitting upstream and then into a Debian kernel. https://patchwork.ozlabs.org/patch/671069/
Thanks Julien, that looks quite interesting. I'll try to remember to look out for this in the mainstream kernels in the future. On Sat, 17 Sep 2016 at 21:56 Julien Goodwin via luv-main < luv-main@luv.asn.au> wrote:
On 16/09/16 18:33, Julien Goodwin via luv-main wrote:
On 16/09/16 11:02, Toby Corkindale via luv-main wrote:
I noticed that Windows 10 now uses CTCP as the default TCP congestion/rate control algorithm, but Linux still defaults to the old Cubic algorithm.
CTCP doesn't appear to be available on Ubuntu LTS at the moment, but there's a whole host of others to choose from. Has anyone here worked out which is the best one to use on typical consumer internet links in Australia?
Over and above the rate control algorithm Linux has a bunch of features that make it work much better than a to-the-spec cubic implementation (not surprising with a bunch of large content providers like $EMPLOYER submitting their fixes upstream).
Things like TCP pacing and the work from the bufferbloat folk have really improved things.
https://fasterdata.es.net/host-tuning/linux/fair-queuing-scheduler/
...and something that finally went public overnight is a new TCP rate control algorithm from some of the folk at $EMPLOYER, I for one am very much looking forward to this hitting upstream and then into a Debian kernel.
https://patchwork.ozlabs.org/patch/671069/
_______________________________________________ luv-main mailing list luv-main@luv.asn.au https://lists.luv.asn.au/cgi-bin/mailman/listinfo/luv-main
On Sat, 17 Sep 2016, Julien Goodwin wrote:
On 16/09/16 18:33, Julien Goodwin via luv-main wrote:
Over and above the rate control algorithm Linux has a bunch of features that make it work much better than a to-the-spec cubic implementation (not surprising with a bunch of large content providers like $EMPLOYER submitting their fixes upstream).
Things like TCP pacing and the work from the bufferbloat folk have really improved things.
https://fasterdata.es.net/host-tuning/linux/fair-queuing-scheduler/
...and something that finally went public overnight is a new TCP rate control algorithm from some of the folk at $EMPLOYER, I for one am very much looking forward to this hitting upstream and then into a Debian kernel.
Damn, that's so obvious it must have taken a genius to think it up :) -- Tim Connors
On Monday, 19 September 2016 11:46:00 AM AEST Tim Connors via luv-main wrote:
Damn, that's so obvious it must have taken a genius to think it up :)
Signed-off-by: Van Jacobson <vanj@google.com> Signed-off-by: Neal Cardwell <ncardwell@google.com> Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Nandita Dukkipati <nanditad@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com> The first line in the signed-off list is noteworthy, someone who didn't get a keynote slot at an LCA... ;) -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
On 19/09/16 15:23, Russell Coker via luv-main wrote:
On Monday, 19 September 2016 11:46:00 AM AEST Tim Connors via luv-main wrote:
Damn, that's so obvious it must have taken a genius to think it up :)
Signed-off-by: Van Jacobson <vanj@google.com> Signed-off-by: Neal Cardwell <ncardwell@google.com> Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Nandita Dukkipati <nanditad@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
The first line in the signed-off list is noteworthy, someone who didn't get a keynote slot at an LCA... ;)
They're all pretty good, Van obviously has a history with TCP, and just a few weeks back Eric & Neal helped me by sending some quick patches upstream for a hard-to-debug issue. (I'm sure the rest are awesome too, I just haven't worked with them)
participants (5)
-
Julien Goodwin -
Julien Goodwin -
Russell Coker -
Tim Connors -
Toby Corkindale