Reliable and Fault-Tolerant Storage Systems - Scalable Continuous Media Streaming Systems - page 92

While summing the previous upper bounds will also bound the request service time, we note
that most, if not all, modern disk drives support a read-on-arrival feature [3] (all five disk drive
models studied in Section 5.7 support read-on-arrival). Specifically, the service time model
assumes that if the disk head arrives to the track to find it passing through the middle of
the media block, it will wait until the first sector of the media block rotates back to the
head position before commencing reading. The read-on-arrival feature removes this restriction
and allows the disk to start reading data immediately even in the middle of the requested
media block. This avoids the worst-case scenario of waiting one complete rotation before
reading.
Therefore, for a service round of k requests, the round length will be bounded from above

by

k

Q
SY min

t hsw + Q

SY min N suf

f seek (1) Q

SY min

W − 1

t max t max

seek

t r ( k ) = α +

seek +

+ +

(5.16)
where the first term represents the worst-case time to read k requests using CSCAN and the
second term is the head-repositioning time.
Reconsider the scheduling algorithm in Section 5.3.2, the server needs to ensure that a
complete parity group must be retrieved within time T r to maintain continuous transmission:

t r ( k ) ≤ T r (5.17)

This timing constraint, also known as the continuity condition, determines the maximum
number of concurrent media streams, denoted by K , that can be supported by the server:

max k

( N D − 1) Q

K = | t r ( k ) ≤ , k = 1 , 2 ,... (5.18)

R v

5.4 Automatic Data Rebuild
A system is said to operate under normal mode when there is no failure. The system switches
to degraded mode of operation once a disk failure occurs. Under this degraded mode of
operation, unavailable data are recomputed in real time from the remaining disks to sustain
service. Although still operational, the system must return to normal mode of operation as
soon as possible because any further disk failure will cripple the entire system. The goal of
data rebuild is to bring the system back to normal mode of operation by reconstructing data
lost in the failed disk into spare storage hardware.
A number of modern disks and disk controllers not only can detect a disk failure, but can
also predict a disk failure in advance. This early-warning signal can be used to initiate data
rebuild even before the actual failure occurs. However, there are also complications that must
be handled properly. First, if the data disks are updated (i.e., being written to) during the
rebuild process, then the spare disk will have to be updated accordingly as well. This is less
of a problem in a media server as the disks primary serve read requests. One can also disallow
updating until the rebuild process completes. Second, if the actual failure occurs in a disk other
than the predicted one, then the rebuild process will have to be aborted and then restarted to