Efficient Server Design for Hybrid Multicast Streaming - Scalable Continuous Media Streaming Systems - page 365

1000

800

600

400

200

0

0 400 800 1200 1600 2000

Server Buffer (in MB)

Our Efficient Server Design (Atlas10k)
Our Efficient Server Design (Barracuda)
Our Efficient Server Design (Cheetah 9LP)
Our Efficient Server Design (IBM18es)
GSS Only (Atlas10k)
GSS Only (Barracuda)
GSS Only (Cheetah 9LP)
GSS Only (IBM18es)

Figure 20.9 Server capacity versus server buffer constraint for four hard disk models

20.5.1 Server Capacity
Figure 20.9 plots the server capacity versus the server buffer size constraint for our efficient
server design and the GSS-based server design for four different disks. Compared toGSS-based
design, our design can increase the server capacity by up to 60%. Moreover, the performance
gain increases to about 40% on average when the buffer available is more than 1GB as ev-
ident in Figure 20.10. This is because the interleaved data placement policy reduces disk
seeks substantially and thus the gain in I/O efficiency due to larger block size becomes more
significant.

20.5.2 Utilization of Disk Capacity
The major advantage of using WSGP is that while achieving full utilization of disk storage
capacity, it provides similar performance improvement as SGP. Results show that with the same
system configuration, SGP can only utilize 91% storage capacity of the disk while WSGP can
achieve 100% utilization.