Вы находитесь на странице: 1из 9

Storage Management

NOOP scheduler
 FIFO + Request Merging.
 Request Merging
 A technique that groups together I/O requests that are physically close
together on the disk
 SSTF !
 reduces average seek time.
 The scheduler assumes I/O performance optimization will be handled at
some other layer of the I/O hierarchy
 by the block device
 by an intelligent HBA such as a Serial Attached SCSI (SAS), RAID
controller or by an externally attached controller
 Potentially can be used for solid state devices.
Deadline Scheduler
Deadline Scheduler
 Guarantee a start service time for each request
 each request has an expiration time / deadline to prevent it starvation.
 Uses three queues:
 sorted queue (scan algo)
 all request sorted to cylinder / track
 read queue (deadline queue)
 read requests sorted based on deadline
 by default, read requests have an expiration time of 500 ms
 write queue (deadline queue)
 by default, write requests have an expiration time of 5 sec
Anticipatory Scheduler
 Scan and Deadline scheduling can be counterproductive
 if there are numerous synchronous read requests
 It seeks to increase the efficiency of HDD by "anticipating" synchronous
read operations.
 Deceptive idleness
 Process is about to issue next request
 Scheduler assumes that process has no further requests!
 More to other area on disk
 After servicing a request
 Wait (Intelligently )
 If a nearby request occurs , service it
 If not … resume deadline scheduling algorithm
 AS was default Linux kernel scheduler between 2.6.0 and 2.6.18
Completely Fair Queuing [CFQ]

RT

10ms Disk Queue


Q1 Dispatcher
Valve

Q2

Disk
Q20
Completely Fair Queuing [CFQ]
 Synchronous requests
 Per process queue
 Fixed time slice for the queue / process to access HDD.
 Length of the time slice and the number of requests a queue is
allowed to submit depends on the IO priority of the given process.
 Asynchronous requests
 All processes are batched together in few queues based on priority.
 CFQ fixes “Deceptive Idealness”
 By batch processing
Memory Hierarchy
Data on DRAM

Вам также может понравиться