From: Tyler Robison (trobison@cs.washington.edu)
Date: Wed Oct 27 2004 - 07:04:26 PDT
This paper describes a way to provide different levels of service
over the existing Internet; that is, although strict performance
guarantees cannot necessarily be made, certain techniques can be used to
allocate bandwidth to users within a given network. Essentially, each
user would have a profile that describes how much bandwidth they are
allowed and what quality of service they should be given, and routers in
the network watch these packets and decide whether they are exceeding
their allocation; if so, the packet is marked as being out of its
allocation, and so if it hits a congested area in the future, that router
can drop the packet instead of other packets that are within their ranges.
The version just described is the sender-based version; for the
receiver-based control version, a packet travels to its destination
without being marked by in/out, and has its ECN bit set if it encounters
congestion along the way. Additionally, the number of 'out' packets
dropped can be an indication of a non-responsive host, and it can be
treated accordingly; so this can be used to help detect non-responsive
hosts as well.
This paper is nice in that it suggests a decent sounding technique
which can be built off the current Internet, which isn't dependent on a
particular application policy (that is, different companies/institutions
could assign bandwidth as they saw fit), and which should be fairly
effective, as suggested by the simulations.
Yet the paper doesn't even mention issues such as security, and
perhaps worse it seems that it would be difficult to implement in
practice; while changes to individual networks wishing to use the services
would be easy enough, some modifications would be needed for every router
on the Internet (or a good number of them, in order for the service to be
effective), since they would need a dropping scheme using the in/out
signals. Even though this implementation wouldn't be difficult, and
wouldn't hurt the utility of the routers, there is also no distinct and
immediate advantage for them, and so it seems unlikely that many would be
updated.
It also seems like there might be easy ways to exploit the
service; if a host in a given network is trying to use too much bandwidth,
its packets will get marked as out and so it will restricted, but if the
network itself decides that all packets will be marked as 'in' (after all,
it would result in better performance for that network), there are no
immediate negative consequences for them.
This archive was generated by hypermail 2.1.6 : Wed Oct 27 2004 - 07:04:27 PDT