Efficient Server Design for Hybrid Multicast Streaming - Scalable Continuous Media Streaming Systems - page 359

and the total server buffer requirement [6] is given by

1
G )

B Server =

+ ,

CNQ (1 (20.3)

where C is the total number of multicast channels in the server.
As the static channels' offset T R is integer multiples of the micro-round length, it is easy
to see that the static channels will be equally distributed to all G groups. The remaining disk
capacity is then used to support dynamic channels.

20.4 An Efficient Server Design
In SS-VoD, there are two types of multicast channels - static and dynamic multicast channel.
Static multicast channels stream the whole video while dynamic channels serve clients with
up to the first T R seconds of the video only. In terms of data access pattern, static channels
retrieve data in a fixed schedule. By contrast, the data access pattern of dynamic channels is
random. In this section, we present an efficient design for the SS-VoD server. Specifically,
this efficient server design has three distinctive features. First, the disk storage is organized
using an improved Weighted Segment Group Pairing (WSGP) scheme to exploit disk zoning
to increase disk throughput and storage utilization. Second, an interleaving data placement
policy is used to store video data to be transmitted over the static channels to exploit the static
channels' periodicity. Third, the first T R seconds of the videos are replicated for placement in
the outermost zones to increase disk throughput for serving the dynamic channels. We design a
new scheduler to schedule the data retrievals for both static and dynamic channels and quantify
its performance. These are presented in details in the following sections.

20.4.1 The Weighted Segment Group Pairing (WSGP) Scheme
Today's hard disks commonly employ zone-bit-recording (ZBR) technique to increase disk
capacity. In zoning, outer tracks are equipped with more sectors than inner tracks to exploit
the increased disk surface area available. With a constant rotation speed, the data transfer rate
of the outer zone is also higher than the inner zones. Traditional deterministic performance
analysis limits one to using the lowest data transfer rate in the innermost zone for system
dimensioning and thus wastes the higher transfer rate available in the outer zones.
To improve disk throughput, we devise a Weighted Segment Group Pairing (WSGP) scheme
based on the Segment Group Pairing (SGP) proposed by Chen and Manu [8]. In WSGP, the
group of data blocks to be retrieved in the same round (i.e., a G j , k ) is divided into two sub-
groups, denoted by G j , k and G j , k . The first group is placed in an outer zone with higher
transfer rate and the second group is placed in an inner zone with lower transfer rate. For a
disk with Z zones, zone h and zone ( Z

1) are paired together and the two sub-groups,
G j , k and G j , k , are allocated to these two zones respectively. In the original SGP [8] the groups
are of equal size. This is undesirable as the zones often have different capacities and the extra
capacity in the larger zone will be wasted. Thus, we extend SGP to WSGP by dividing the
group into sub-groups of sizes proportional to the zone capacities. Results show that with
WSGP, we can achieve 100% disk storage utilization while achieving the same throughput as
SGP.

− h +