Information Technology Reference
In-Depth Information
Second, the processing delay of bandwidth adjustment may introduce a subtle problem.
Regardless of the resource reservation protocols adopted, a sender (e.g., a video server) that
desires to adjust a connection's bandwidth must first send protocol messages to one or more
network controllers (e.g., routers). The network controllers may in turn need to contact other
controllers along the path of the connection before the request can be granted or denied. In any
case, this process will take time and the time it takes will depend on a lot of factors, such as
the network topology, the reservation protocol adopted, the current utilization of the network,
the number of resource-reservation requests being processed, loss of control messages, etc.
The point is, not only the processing itself takes time, the time it takes also varies. This creates
another problem in upward bit-rate adjustments as delay or even transmission losses may
occur if an adjustment cannot be completed in time. Conceivably one can issue the upward
adjustments well ahead of time to prevent delay/loss but estimating the correct lead-time is by
no means trivial.
We address the two previously discussed problems in this chapter by developing a scheduler
for transmitting VBR videos that can provide deterministic performance guarantee in a mixed-
traffic network and is immune to random delays in processing network resource reservation
requests. The principle of the scheduler, called Monotonic Decreasing Rate (MDR) scheduler,
is to eliminate upward bit-rate adjustments altogether. That is, the transmission schedule is
composed of a series of segments, of which each segment is assigned a bit-rate strictly lower
than the previous segment. Now without the need for upward bit-rate adjustment, resource
reservations are guaranteed to be successful. Moreover, the timing of the bit-rate adjustments
is no longer critical as video data transmission will not be affected by a later-than-expected
downward bit-rate adjustment.
Intuitively, one will expect the MDR scheduler to require more client buffer as video data
are transmitted more aggressively than other smoothing algorithms. Using real-world VBR
video bit-rate traces, we quantify the trade-off and show that for some video streams, the
buffer requirement is indeed increased when compared to smoothing algorithms. To tackle this
problem, we develop an Aggregated Monotonic Decreasing Rate (AMDR) scheduler to enable
one to control the buffer requirement to the same level as smoothing algorithms. Surprisingly,
simulation results show that the AMDR scheduler can achieve performance comparable to
existing smoothing algorithms even when equipped with the same buffer requirement. Thus,
using the AMDR scheduler, one can provide performance guarantee in streaming VBR videos
over mixed-traffic networks with no trade-off in terms of admission complexity, network
utilization, client waiting time, and client buffer requirement.
7.3 Monotonic Decreasing Rate Scheduler
As discussed in Section 7.2, the fundamental limitation of existing smoothing algorithms is
the need for upward rate adjustments, of which correct operation depends on the successful
and timely completion of network resource reservations. To remove this limitation, we can
use only downward rate adjustments in the transmission schedule. In other words, the initial
transmission rate will be the highest, with each subsequent rate lower than the previous one.
We call this algorithm the monotonic decreasing rate (MDR) scheduler for obvious reasons.
In this chapter, we focus on prerecorded videos. The MDR schedule for a video is computed
offline and is stored with the video for use during video streaming. We present an algorithm to
compute the MDR schedule in the next section and derive several properties of the algorithm
in Section 7.3.2 to Section 7.3.4.
Search WWH ::




Custom Search