Efficient Server Design for Hybrid Multicast Streaming - Scalable Continuous Media Streaming Systems - page 357

storage capacity in the outer zones becomes unused. To tackle this limitation, we extend the
SGP policy to account for zone capacity differences. This new policy, calledWeighted Segment
Group Pairing (WSGP), allocates data blocks to the inner and outer zones in proportion to the
zone capacities, thus eliminating the above-mentioned limitation.
Nevertheless, the WSGP policy still does not address the problem of supporting both ran-
dom and periodic video streams. In our server design, we develop a new placement pol-
icy and I/O scheduler incorporating the virtues of CSCAN/GSS for random video streams,
and data interleaving for periodic video streams. In the next section, we first devise a GSS-
based server design and then introduce our new server design in Section 20.4. We then
compare their performances using numerical results in Section 20.5 and give a summary in
Section 20.6.

20.3 A GSS-based Server Design
In this section, we apply the well-known Grouped Sweeping Scheme (GSS) scheduler for use
in a SS-VoD server. This design will serve as a baseline to compare the efficient server design
to be presented in Section 20.4.
Let N be the number of disks in the system, assuming the disks are homogeneous. The
disk's storage is divided into fixed-size blocks of Q bytes each, and a service group is defined
to consist of all the data blocks at the same location from each of the N disks.
Video data are striped across the N disks as shown in Figure 20.1, effectively forming a
RAID-4 [10] disk array without parity. Denote the j th data block of video i by b i , j . Then, the
first N blocks of video i ,[ b i , 1 , b i , N ] are allocated to the first service group. This

storage allocation scheme ensures load balance among all N disks.
Figure 20.2 depicts the Grouped Sweeping Scheme (GSS). In GSS, a macro-round is divided
evenly into G micro-rounds, with each micro-round serving a separate group of video streams.
Assuming all the videos are encoded using constant-bit-rate (CBR) encoding method with the
same bit-rate R V , then in each micro-round the server retrieves one data block from each disk
for each channel, and this data block is then multicast over the next G micro-rounds (i.e., one

b i , 2 , b i , 3 ...

b i,1 b i,1 b i,2 b i,2 b i,3 b i,,3 b i,N- b i,N−2 b i, b i,N−1 b i,N b i,N

… …

1

b i,N b i,N+1 b i, b i,N+2 b i, b i,N+3 b i,2 b i,2N−2 b i,2 b i,2N−1 b i,2N b i,2N

… …

2 3 2 1

b i, b i,2N+1 b i,2N+2 b i,2N+2 b i, b i,2N+3 b i,3 b i,3N−2 b i,3N-1 b i,3N−1 b i,3N b i,3N

… …

+1 +3 2

… … Disk N−2 Disk N−1 Disk N

Disk 1 Disk 2 Disk 3

Figure 20.1 Allocation of video blocks among disks for video i