Transcript Document

Wresting Control from BGP: Scalable Fine-grained Route Control
Patrick Verkaik, Dan Pei, Tom Scholl, Aman Shaikh , Alex C. Snoeren, Kobus van der Merwe
UCSD / AT&T Labs
contact: [email protected]
Motivation:BGP causes load imbalance
Routers
ISP
Egres
s
routes
Customer C
Traffic to
customer
Routes for customer
Rest of the Internet
Typical BGP routing
solution, hot-potato routing:
2. IRSCP receives all egress
1. Route
control
application
(Intelligent Route Service
Control
Point)
routes from nearby customers
measures the network and
and IRSCPs
sends egress lists to
ISP
congestion!
4.
C
•Routers pick nearest egress
route
•Leads to load imbalance for C
•Potentially
yielding
congestion
•BGP protocol controls ISP’s Below
routing
solution
fixes the
routing:
problem but unachievable in BGP
•BGP advertises C to routers
•Each router runs BGP
decision process to select
route to C
ISP
•Traffic follows reverse path
along selected routes
C
Performance of prototype IRSCP
•
•
•
•
•
We propose: route control using
IRSCP
1.
App
IRSCP
4.
Achieves this
routing solution
IRSCP
IRSCP
2.
2.
4.
4.
IRSCP that avoid load
Egress list for these three routers
imbalances
#1
#2
#1
#2
3. IRSCP runs decision process: for Egress list contains egress routes
ordered
by
preference
each router, IRSCP decision process
4. IRSCP Informs nearby
picks most preferred (according to
routers of its decision
egress list) available egress route
Achieved output rate
Estimated max
required input
rate
Connect to 40 emulated IRSCPs and 255 emulated (customer and ISP) routers
Vary number of ISP routers and plot maximum sustained throughput
Due to per-ISP-router decision process, more ISP routers means less throughput
Estimated 95
perc. required
IRSCP easily keeps up with estimated required 95-percentile input rate
To do: overcome bursts of maximum required input rate using flow controlinput rate
Achieved input rate