A Pruning-Based Disk Scheduling Algorithm for Heterogeneous I/O Workloads

Table 1

A summary of disk scheduling algorithms.

Algorithm	Basic idea	Advantage	Weakness	Target applications
SSTF	Service request with shortest seek time first	Simple to implement; high throughput	High variation of response time	Best-effort applications
SATF	Service request with shortest access time (including rotational latency) first	High throughput	Require knowledge of disk structure
SCAN	Scan in one direction and service all requests by track number order and change the direction of scan	Simple to implement; high throughput	Consider only best-effort requests
C-SCAN	Variant of SCAN that always scans in one direction	Simple to implement; high throughput; low variation of response time	Consider only best-effort requests
EDF	Service request with Earliest Deadline First	Simplest to implement in real-time environment	Low disk utilization	Real-time applications
SCAN-EDF	Service EDF order and use SCAN as a tie breaker	Simple to implement	Possible to degenerate into EDF
SSEDO/; SSEDV	Consider both deadline and seek time, but put more weight on deadline	Consider both deadline and seek time	Require parameter tuning
FD-SCAN	Move head towards the request with earliest feasible deadline; service requests on the way	Consider feasibility of real-time requests	May incur many deadline misses; high overhead
SCAN-RT	Basically SCAN; insert new request only if it does not violate the deadlines of pending requests	Employ SCAN considering deadline	Not consider different priority level of requests
DM-SCAN	Apply SCAN by unifying deadlines of requests within maximum scannable group	Employ SCAN considering deadline	Possible to degenerate into EDF
Kamel’s	Basically SCAN; insert new request considering deadline and priority	Consider different priority level; deadline guarantee	Immature handling of requests in next round
MS-EDF	Find a global optimal schedule using branch-and-bound scheme	Global search of an optimal schedule; high performance	Only for real-time requests; off-line mechanism
Rangan’s	A fixed-order cyclical scheduling strategy	Employ an elaborate disk model	Not handle frame-oriented data	Multimedia streaming applications
GSS	Assign the joint deadline to each group of streams; each group is serviced in a fixed order in a round	Simple to implement; obtain high throughput by using SCAN within each round	Require group size tuning
Preseeking sweep	Split stream data requests into multiple fragments	Obtain high throughput by employing an elaborate disk model	Require knowledge of disk structure
Chen’s	Modify round-robin scheduling to provide statistical guarantees to clients	Useful when playback guarantee is not necessary	Require complicated statistical analysis
Cello	Two-level disk scheduling framework: a class-independent scheduler and a set of class-specific schedulers	Guarantee predefined disk bandwidth for each class	Not guarantee the jitter-free playback of multimedia	Applications with heterogeneous workloads
Reddy’s	Similar to Cello; employ admission controller as well as scheduler	Consider admission controller and VBR streams	Require knowledge of disk structure
Won’s	Allocate some bandwidth to best-effort requests by extending the length of round	Consider buffer requirement for jitter-free playback of multimedia	Require knowledge of disk structure
WRR-SCAN	Allocate disk bandwidth to prioritized task groups and service requests in the group by SCAN	Guarantee minimal disk bandwidth for aperiodic tasks	Require knowledge of disk structure
Cascaded-SFC	Unified framework considering various scheduling parameters	Consider all scheduling parameters; applicable to various environments	Require parameter tuning
Fahrrad	Reserve disk bandwidth based on disk time utilization	Fully reserve the disk bandwidth for different applications	Less efficient in small bursty workload with low-latency targets
Horizon	Two-layered approach: upper level for deadline assignment and lower level for scheduling	Schedule requests based on their expected disk service time	Lack of hard real-time supports