bbr vs cubic network performance on high speed rail networks

144 views
Skip to first unread message

Dave Taht

unread,
Dec 17, 2018, 10:42:33 PM12/17/18
to bloat, BBR Development
It's really weird to worry about throughput and latency at 300km/h vs
350km/hr, being as the fastest US train I've ever been on rarely
cracks 100kph.

https://arxiv.org/pdf/1812.04823.pdf - the handover bit (section 5.2
and later) was pretty interesting.

Round-trip-time (RTT). As shown in Fig. 6, BBR has more than twice
lower RTTs than CUBIC (e.g., 191.53 ms versus 431.35 ms at 300 km/h,
and 148.63 ms versus 345.02 ms at 350 km/h for median value) due to
their different CCA design rationales: BBR intends to suppress the RTT
to overcome the bufferbloat problem [24


--

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-205-9740

Tong

unread,
Dec 19, 2018, 9:45:34 PM12/19/18
to BBR Development
HAHA,  I think the authors have taken measurements in China, where most of the high speed rail train has the peak speed of 310 km/h. So actually it is weird to say "it is wired to evaluate TCP at 300+ kmph because US train does not reach 300+kmph"...

在 2018年12月18日星期二 UTC+8上午11:42:33,Dave Taht写道:

Jonathan Morton

unread,
Dec 20, 2018, 1:06:19 AM12/20/18
to Tong, BBR Development
> On 20 Dec, 2018, at 4:45 am, Tong <super...@gmail.com> wrote:
>
> HAHA, I think the authors have taken measurements in China, where most of the high speed rail train has the peak speed of 310 km/h. So actually it is weird to say "it is wired to evaluate TCP at 300+ kmph because US train does not reach 300+kmph"...

Japan's Shinkansen network regularly reaches 300-320kph as well, and they have at least had prototypes running at 360kph - not sure about in-service speeds yet. The constraint for them is a requirement to bring the train to a halt within the earthquake warning time, which requires both aerodynamic spoilers and single-use emergency friction brakes; otherwise, some of their lines are already built for the higher speed.

The principal lines of Britain's intercity network support speeds of 200kph, or at least something reasonably close to that. On the WCML, this applies only to tilting trains, but on the GWML and ECML this speed is available to conventional trains with sufficiently good brakes and suspension. A true high-speed network for 300kph+ capability is in the early construction phases, ignoring HS1 which is really an extension of the French railway through the Channel Tunnel; the principal holdups here are mostly NIMBYism.

Continental Europe has an extensive high-speed railway network as well. I believe some TGV lines in France regularly run at 360kph now, while 320kph is more common in Germany. They set a record of over 500kph with a test train, which also required the section of railway concerned to be specially modified (extra tension in the overhead wires).

All of this flatly ignores maglevs, which are so ludicrously expensive to build that only Japan is seriously embarking on an intercity installation (the Chūō Shinkansen), due solely to lack of capacity on the Tokaido Shinkansen already connecting the same cities. I understand 500kph+ is targeted.

In comparison to all this, American railways are bizarrely backwards. I think this is largely because major infrastructure has been left to the free market - but with the weird exception of the highways, which more-or-less directly led to the 1970s oil crisis and some of today's environmental problems.

- Jonathan Morton

Reply all
Reply to author
Forward
0 new messages