Continuous Media Storage and Retrieval - Scalable Continuous Media Streaming Systems - page 45

where n is the number of tracks to seek, is the

proportionality constant for seek time. Seek time is proportional to the square root of n to
approximate the effect of physically accelerating the disk head.

α is the parameter for fixed overhead, and β

3.2 Disk Scheduling
The previous disk model quantifies the time to retrieve one block of media data. Amedia server
often serves multiple media streams concurrently, with each stream issuing retrieval requests
periodically. In the simplest case, the system can process retrieval requests in the order they
are issued, i.e., first-come-first-serve (FCFS) scheduling. This simple disk scheduler, however,
can incur excessive overheads in seeking, as illustrated in Figure 3.2.
In this example three retrieval requests identified as 1, 2, and 3 are issued in that order by
on-going media streams. The data for streams 1 and 3 are located in the outer tracks while
the data for stream 2 are located in the inner track. As a result, the disk head will need to first
travel to the outer track to retrieve data for stream 1, and then travel all the way across the disk
surface to the inner track to retrieve data for stream 2, and finally travel all the way back to the
outer track to retrieve data for stream 3. In the worst case the service time to retrieve one data
block becomes

T fcfs = T read ( N track − 1) (3.2)

This is extremely inefficient as the data for stream 1 and 3 are in fact located in nearby tracks.
This contrived example illustrates the inefficiency of the FCFS scheduler, which is seldom
employed in media servers.
Most media servers instead employ round-based disk schedulers to reduce the disk seek
overhead. Figure 3.3 depicts a common retrieval and transmission scheduler serving requests
frommultiple concurrent media streams, denoted by numeric 1 to 5. For simplicity, we assume
the media streams are homogeneous and have the same average data rate, denoted by R .
Therefore, in the simplest case, the scheduler will retrieve one fixed-size block of data, say, of
Q bytes, for each of the active streams in a service round of duration Q / R seconds.

Disk
Head

2 3 1

Disk Platter

1 2 3

Request arrival order:

1 2 3

Request service order:

Figure 3.2 First-come-first-serve scheduling of disk retrieval requests