- Consider the following network with 2 flows:
- Source B
transmits a
massive amount of data
to destination D
- Suppose source A transmits a packet
P
at time
t = t0
to destination C
- $64,000 question:
-
Is there some way to
guarantee
that the packet
P
will arrive at destination C
within a certain amount of time???
- It turns out that the
Exponential Server
can solve the
delay guarantee problem
- Consider the unweighted
Exponential Server :
- The delay
in transmission of a packet of a flow f
depends LARGELY on:
-
The amount of data
sent prior to that packet
by flow f
- This fact is caused by the way that the
Exponential Server
services the packets of the active flows.
Recall how an Exponential Server operates:
- Suppose a processor has the
capability of processing
M
units of work per sec.
- If there is
only one job ,
then this job is worked on by the processor
at
a speed of M units/sec
- On the other hand, if there is
two jobs available,
then
both jobs are worked on simultaneously
by the processor at a
speed of M/2 units/sec
- And if there is
three jobs available,
then
all 3 jobs are worked on simultaneously
by the processor at a
speed of M/3 units/sec
- And so on...
- Suppose a processor has the
capability of processing
M
units of work per sec.
- So the packets of a flow
f
will be worked on (transmitted) by the
Exponential Server - albeit at a lower data rate.
- But
because you are
always making progress in the transmission,
the Exponential Server
can
make certain
guarantees
- As you know, the Exponential Server cannot be implemented
in real system, and we have studied many approximation
algorithms.
- Because these various algorithm
approximates the Exponential Server
that can provide delay guarantees,
the various approximation algorithms
will also be able to
provide delay guarantees
(albeit a "lower grade" guarantee than the Exponential Server
- We will next study a seminal paper on the level of delay guarantee that each approximation algorithm (WFQ, SCFQ and BSFQ) can provide.