Unformatted text preview: ile our initial simulations with large latencies did not reveal any signi cant
problems, we do not yet understand CSFQ well enough to
be con dent in the viability of this all-core" design. However, if viable, this all-core" design would allow all interior
routers to have only very simple forwarding and dropping
mechanisms, without any need to classify packets into ows.
In addition, we should note that it is possible to use
a CSFQ-like architecture to provide service guarantees. A
possible approach would be to use the route pinning mechanisms described in 23 , and to shape the aggregate guaranteed tra c at each output link of core routers 6 .
One of the initial assumptions of this paper was that
the more traditional mechanisms used to achieve fair allocations, such as Fair Queueing or FRED, were too complex
to implement cost-e ectively at su ciently high speeds. If
this is the case, then a more scalable approach like CSFQ
is necessary to achieve fair allocations. The CSFQ islands
would be comprised of high-speed backbones, and the edge
routers would be at lower speeds where classi cation and
other per- ow operations were not a problem. However,
CSFQ may still play a role even if router technology advances to the stage where the more traditional mechanisms
can reach su ciently high speeds. Because the core-version
of CSFQ could presumably be retro t on a sizable fraction of the installed router base since its complexity is roughly
comparable to RED and can be implemented in software,
it may be that CSFQ islands are not high-speed backbones
but rather are comprised of legacy routers.
Lastly, we should note that the CSFQ approach requires
some con guration, with edge routers distinguished from
core routers. Moreover, CSFQ must be adopted an island
at a time rather than router-by-router. We do not know if
this presents a serious impediment to CSFQ's adoption. References
1 J.C.R. Bennett, D.C. Stephens, and H. Zhang. High speed,
scalable, and accurate implementation of...
View Full Document
- Fall '10
- Eugene Ng
- Scheduling algorithm, Round-robin scheduling, Scheduling algorithms, Ow, Fair queuing, ows