Disk Storage and Dependability 8'1, 8'2

1
Disk Storage and Dependability8.1, 8.2

• John Ashman

2
Introduction

• Users get frustrated when a computer
  crashes/hangs and needs rebooting they get livid
  when they lose saved memory.
• I/O systems place greater emphasis on
  dependability and cost, while processors and
  memory emphasize performance and cost.

3
I/O Performance

• Some systems may require greater throughput
  others may care about access latency.
• Performance depends on such things as device
  characteristics, the devices connection to the
  system, memory, or the OS.

4
I/O Devices
5
I/O Characteristics

• Behavior
• Input, Output, or Storage.
• Partner
• Human or machine on opposite end, writing or
  reading data.
• Data Rate
• Peak rate at which data can be transferred
  between I/O and main memory/processor.

6
Measuring Performance

• 1. How much data can we move through the data in
  a given time?
• 2. How many I/O operations can we do per unit of
  time?
• Example The NITS cares about processing many tax
  forms, stored in separate, small files. An I/O
  supporting simultaneous transfer of many small
  files is preferable.

7
I/O Requests

• The number of reads or writes to I/O devices.
• Desktops/Laptops greater response time.
• Servers throughput and expandability.
• Some systems require both. Ex ATMs need to be
  able to quickly process transactions process
  multiple at a time.

8
Disk Storage

• Magnetic disks a rotating platter coated with a
  magnetic surface.
• Nonvolatile data remains when powered down.
• Tracks concentric circles the disk is divided
  into (10k to 50k typically).
• Sectors divisions of tracks (100 to 500/track,
  512 bytes moving to 4096 bytes).
• Zone Bit Recording varying number of sectors
  per track.
• Cylinder all tracks under the heads at a time.

9
Accessing Data

• Seek and Seek Time positioning the head over
  the correct track and the time it takes,
  respectively.
• Manufacturers report min, max, and average seek
  times. Average time for all possible seeks /
  number of possible seeks.
• Actual times may be only 25 - 33 of reported
  times, due to locality.

10
Rotational Latency/Delay

• Time taken for the designated sector to rotate
  under the head. Average latency is halfway
  around the disk.
• ARL 0.5 / ( 5400 x 60) 5.6 ms
• ARL 0.5 / ( 15000 x 60) 2.0 ms
• Average rotational latency is between 5.6 ms and
  2.0 ms.

11
Transfer Time

• This is the time it takes to transfer a block of
  bits.
• A function of sector size, rotation speed, and
  the recording density of the track.
• A disk controller handles the disk and transfer
  between memory.
• Controller time is the last variable it is the
  overhead the controller imposes on an I/O access.

12

• What is the average time to read or write a
  512-byte sector for a typical disk rotating at
  10,000 RPM? The advertised average seek time is 6
  ms, the transfer rate is 50 MB/sec, and the
  controller overhead is 0.2 ms.
• Average disk access time Average seek time
  average rotational time transfer time
  controller overhead
• 6.0 ms .5 rotation/ 10,000 RPM 0.5 KB / 50
  MB/s 0.2 ms 6.0 3.0 0.01 0.2
  9.2 ms

13
Dependability, Reliability, Availability

• 1. Service Accomplishment the service is
  delivered as specified.
• 2. Service Interruption the delivered services
  is different than the requested one.
• 1 ? 2 is a failure. 2 ? 1 is a restoration.
• Failures can be either permanent or intermittent.

14
Reliability and Availability

• Reliability a measure of the continuous service
  accomplishment mean time to failure.
• Availability a measure of service
  accomplishment with respect between
  accomplishment and interruption.
• Availability MTTF / (MTTFMTTR)
• MTTF / (MTBF)

15
Hard Drive Example

• Visuals
• Figure 8.3 on page 572
• Specs
• Figure 8.4 on page 573
• Causes of Failure
• Figure 8.5 on page 574

16
Increasing MTTF

• 1. Fault Avoidance preventing fault occurrence
  by construction.
• 2. Fault Tolerance redundancy used to allow
  services to comply with the specification despite
  a fault. (hardware)
• 3. Fault Forecasting predicting the
  presence/creation of faults. (hardware and
  software)

17
RAID

• Redundant Arrays of Inexpensive Disks
• Basically, system(s) of using multiple hard disks
  to increase both performance and reliability of
  storage.

18
Raid 0 No Redundancy

• Stripes data across two or more hard disks,
  improving performance through the use of multiple
  heads pseudo-RAID.

19
RAID 1 - Mirroring

• This requires twice as many disks as RAID 0 data
  written to one disk is mirrored on another. The
  most expensive RAID setup.

20
RAID 3 Bit-Interleaved Parity

• The cost of availability is lessened to 1/N,
  where N is the number of disks in a protection
  group.
• Protection Group disks that share a common
  check disk or block.
• Information lost on a disk can be restored from
  the remaining disks.
• Parity is the process in which all good data left
  on working disks is subtracted from the parity
  disk the result is the missing data.

21
RAID 4 Block-Interleaved Storage

• RAID 4 watches the bits of information being
  changed, rather than comparing each remaining
  disk.
• This requires a mixture of large reads, large
  writes, small reads, and small writes.

22
RAID 3 vs. RAID 4
23
RAID 5 Distributed Block-Interleaved Parity

• The parity is spread across multiple disks,
  rather than a single parity disk.
• Rather than bottlenecking when two consecutive
  writes are queued, this limits bottlenecking only
  to certain situations.

24
RAID 4 vs. RAID 5
25
RAID 6 PQ Redundancy

• This is a rarely used configuration that allows
  for the user to safeguard against multiple disk
  failure.
• The overhead is twice that of RAID 5
• P and Q are both parity tracks, require an
  increased number of disk accesses for recovery
  and comparison.

26
RAIDs in a Nutshell

• 80 of servers use some RAID 1-5.
• Hot swapping disks is required to make repairs
  the system cannot be shutdown.
• Environmental damage often affects all the disks
  in the system.
• Human error is not incorporated into MTTF.
• RAID 6 is good for the worriers the extra cost
  may pay off in protection from human error and
  environmental hazard.