[Bloat] bufferbloat at 10 and 100gig

[Dave Taht <dave.taht@gmail.com>]

http://www.ece.virginia.edu/~mv/PhDthesis/zhenzhen/zz-dissertation.pdf

showed 60ms of bloat in the 10gig data. And while I do love the idea of
having 100gbit national infrastructure, I do hope that they are using newer
versions of linux than 2.6.33 to develop it with.

See pages 24

and

experimental network setup is shown in Figure 4.1. It was called the
Long Island MAN
(LIMAN) testbed, and was supported by ESnet as a DOE-funded testbed
for networking
research. The high-performance hosts,
W1
(West 1),
E1
(East 1), and
E2
(East 2), were Intel
Xeon Nehalem E5530 models (2.4GHz CPU, 24GB memory) and ran Linux
version 2.6.33.
The application hosts,
WA
(West App-host) and
EA
(East App-host), were Intel Dual 2.5GHz
Xeon model and ran Linux 2.6.18. The routers,
WR
(West Router) and
ER
(East Router),
were Juniper MX80's running Junos version 10.2. The link rates were 10
Gbps from the
high-performance hosts to the routers, 1 Gbps from the application
hosts to the routers,
and 10 Gbps between the routers
-- 
Dave Täht

Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html