INF5070 – Media Servers and Distribution Systems: Storage ...€¦ · INF5070 – media servers...
Transcript of INF5070 – Media Servers and Distribution Systems: Storage ...€¦ · INF5070 – media servers...
Storage Systems Storage Systems ––Part IPart I
3/10 - 2005
INF5070 – Media Servers and Distribution Systems:
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
OverviewDisks
mechanics and properties
Disk schedulingtraditionalreal-timestream orientedmixed media
Data placement
Complicating factors
Disks
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
DisksDisks are orders of magnitude slower than main memory, but are cheaper and have more capacity
Disks are used to have a persistent system and manage huge amounts of information
Because......there is a large speed mismatch compared to main
memory (this gap will increase according to Moore’s law), ...disk I/O is often the main performance bottleneck...we need to minimize the number of accesses,...
...we must look closer on how to manage disks
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disks
Two resources of importance
storage space
I/O bandwidth
Several approaches to manage multimedia data on disks:
specific disk scheduling and large buffers
optimize data placement for continuous media
replication / striping
combinations of the above
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Mechanics of Disks
Platterscircular platters covered with magnetic material to provide nonvolatile storage of bits
Tracksconcentric circles on asingle platter
Sectorssegment of the track circle –usually each contains 512 bytes –separated by non-magnetic gaps.The gaps are often used to identifybeginning of a sector
Cylinderscorresponding tracks on the different platters are said to form a cylinder
Spindleof which the platters rotate around
Disk headsread or alter the magnetism (bits) passing under it. The heads are attached to an arm enabling it to move across the platter surface
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk SpecificationsSome existing (Seagate) disks today:
Note 1:disk manufacturers usuallydenote GB as 109 whereascomputer quantities often arepowers of 2, i.e., GB is 230
Note 3:there is usually a trade off between speed and capacity
Note 2:there is a difference between internal and formatted transfer rate. Internalis only between platter. Formatted is after the signals interfere with the electronics (cabling loss, interference, retransmissions, checksums, etc.)
8 MB4 MB16 MBdisk buffer cache
609 – 891 520 – 682282 – 508 internal transfer rate (Mbps)
234.17average latency
71216max (full stroke) seek (ms)
0.20.60.8min (track-to-track) seek (ms)
3.6 5.77.4average seek time (ms)
18.4799.77224.247#cylinders
15.00010.0007200Spindle speed (RPM)
73.436.4181.6Capacity (GB)
Cheetah X15.3Cheetah 36Barracuda 180
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk CapacityThe size of the disk is dependent on
the number of platters whether the platters use one or both sidesnumber of tracks per surface(average) number of sectors per tracknumber of bytes per sector
Example (Cheetah X15):4 platters using both sides: 8 surfaces18497 tracks per surface617 sectors per track (average)512 bytes per sectorTotal capacity = 8 x 18497 x 617 x 512 ≈ 4.6 x 1010 = 42.8 GBFormatted capacity = 36.7 GB
Note 1:the tracks on the edge of the platter is larger than the tracks close to the spindle. Today, most disks are zoned, i.e., the outer tracks have more sectors than the inner tracks
Note 2:there is a difference between formatted and total capacity. Some of the capacity is used for storing checksums, spare tracks, gaps, etc.
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Access Time
How do we retrieve data from disk?position head over the cylinder (track) on which the block (consisting of one or more sectors) are locatedread or write the data block as the sectors move under the head when the platters rotate
The time between the moment issuing a disk request and the time the block is resident in memory is called disk latency or disk access time
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Access Time
+ Rotational delay
+ Transfer time
Seek time
Disk access time =
+ Other delays
Disk platter
Disk arm
Disk head
block xin memory
I wantblock X
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Access Time: Seek TimeSeek time is the time to position the head
the heads require a minimum amount of time to start and stop moving the headsome time is used for actually moving the head –roughly proportional to the number of cylinders traveled
Time to move head:
~ 10x - 20x
x
1 NCylinders Traveled
Time
“Typical” average: 10 ms → 40 ms7.4 ms (Barracuda 180)5.7 ms (Cheetah 36)3.6 ms (Cheetah X15)
nβα + number of tracksseek time constantfixed overhead
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Access Time: Rotational Delay
Time for the disk platters to rotate so the first of the required sectors are under the disk head
head here
block I want
Average delay is 1/2 revolution
“Typical” average: 8.33 ms (3.600 RPM)5.56 ms (5.400 RPM)4.17 ms (7.200 RPM)3.00 ms (10.000 RPM)2.00 ms (15.000 RPM)
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Access Time: Transfer TimeTime for data to be read by the disk head, i.e., time it takes the sectors of the requested block to rotate under the head
Transfer rate =
Transfer time = amount of data to read / transfer rate
Example – Barracuda 180:406 KB per track x 7.200 RPM ≈ 47.58 MB/sExample – Cheetah X15:316 KB per track x 15.000 RPM ≈ 77.15 MB/s
Transfer time is dependent on data density and rotation speedIf we have to change track, time must also be added for moving the head
amount of data per tracktime per rotation
Note:one might achieve these transfer rates reading continuously on disk, but time must be added for seeks, etc.
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Access Time: Other Delays
There are several other factors which might introduce additional delays:
CPU time to issue and process I/Ocontention for controllercontention for buscontention for memoryverifying block correctness with checksums (retransmissions)waiting in scheduling queue...
Typical values: “0”(maybe except from waiting in the queue)
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk ThroughputHow much data can we retrieve per second?
Throughput =
Example:for each operation we have
- average seek - average rotational delay- transfer time - no gaps, etc.
Cheetah X15 (max 77.15 MB/s)4 KB blocks 0.71 MB/s64 KB blocks 11.42 MB/s
Barracuda 180 (max 47.58 MB/s)4 KB blocks 0.35 MB/s64 KB blocks 5.53 MB/s
data sizetransfer time (including all)
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Block Size The block size may have large effects on performanceExample:assume random block placement on disk and sequential file access
doubling block size will halve the number of disk accesseseach access take some more time to transfer the data, but the total transfer time is the same (i.e., more data per request)halve the seek times halve rotational delays are omitted
e.g., when increasing block size from 2 KB to 4 KB (no gaps,...)for Cheetah X15 typically an average of:☺ 3.6 ms is saved for seek time☺ 2 ms is saved in rotational delays
0.026 ms is added per transfer time
increasing from 2 KB to 64 KB saves ~96,4 % when reading 64 KB
} saving a total of 5.6 ms when reading 4 KB (49,8 %)
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
100 200 300 400 5000
10
20
30
40
50
60
70
80
90
100
Amount of data read per operation (KB)E
ffic
iency in %
of m
ax. th
roughput
Block SizeThus, increasing block size can increase performance by reducing seek times and rotational delays(figure shows calculation on some older device)
But, blocks spanning several tracks still introduce latencies…
… and a large block size is not always best
small data elements may occupy only a fraction of the block (fragmentation)
Which block size to use therefore depends on data size and data reference patterns
The trend, however, is to use large block sizes as new technologies appear with increased performance – at least in high data rate systems
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Writing and Modifying BlocksA write operation is analogous to read operations
must add time for block allocation
a write operation may has to be verified – must wait another rotation and then read the block to see if it is the block we wanted to write
Total write time ≈ read time (+ time for one rotation)
Cannot modify a block directly:read block into main memory
modify the block
write new content back to disk
(verify the write operation)
Total modify time ≈ read time + time to modify + write time
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk ControllersTo manage the different parts of the disk, we use a disk controller, which is a small processor capable of:
controlling the actuator moving the head to the desired track
selecting which platter and surface to use
knowing when right sector is under the head
transferring data between main memory and disk
New controllers acts like small computers themselves
both disk and controller now has an own buffer reducing disk access time
data on damaged disk blocks/sectors are just moved to spare room at the disk – the system above (OS) does not know this, i.e., a block may lie elsewhere than the OS thinks
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Efficient Secondary Storage UsageMust take into account the use of secondary storage
there are large access time gaps, i.e., a disk access will probably dominate the total execution timethere may be huge performance improvements if we reduce the number of disk accessesa “slow” algorithm with few disk accesses will probably outperform a “fast” algorithm with many disk accesses
Several ways to optimize .....block size - 4 KB file management / data placement - variousdisk scheduling - SCAN derivatemultiple disks - a specific RAID levelprefetching - read-ahead prefetchingmemory caching /replacement algorithms - LRU variant…
Disk Scheduling
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Scheduling – ISeek time is the dominant factor of total disk I/O time
Let operating system or disk controller choose which request to serve next depending on the head’s current position and requested block’s position on disk (disk scheduling)
Note that disk scheduling ≠ CPU schedulinga mechanical device – hard to determine (accurate) access timesdisk accesses cannot be preempted – runs until it finishesdisk I/O often the main performance bottleneck
General goalsshort response timehigh overall throughput fairness (equal probability for all blocks to be accessed in the same time)
Tradeoff: seek and rotational delay vs. maximum response time
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Scheduling – II
Several traditional (performance oriented) algorithmsFirst-Come-First-Serve (FCFS)Shortest Seek Time First (SSTF)SCAN (and variations)Look (and variations)…
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
First–Come–First–Serve (FCFS)FCFS serves the first arriving request first:
Long seeks“Short” average response time
time
cylinder number1 5 10 15 20 25
12
incoming requests (in order of arrival):
14 2 7 21 8 24
schedulingqueue
248
217
2
14
12
Note:the lines only indicate some time – not exact amount
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Shortest Seek Time First (SSTF)SSTF serves closest request first:
short seek timeslonger maximum seek times – may even lead to starvation
time
cylinder number1 5 10 15 20 25
12
incoming requests (in order of arrival):
14 2 7 21 8 24
schedulingqueue
24821721412
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
SCANSCAN (elevator) moves head edge to edge and serves requests on the way:
bi-directionalcompromise between response time and seek time optimizations
time
cylinder number1 5 10 15 20 25
12
incoming requests (in order of arrival):
14 2 7 21 8 24
schedulingqueue
24821721412
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
SCAN vs. FCFS
Disk scheduling makes a difference!
In this case, we see that SCAN requires much less head movement compared to FCFS(here 37 vs. 75 tracks)
cylinder number1 5 10 15 20 25
time
time
12incoming requests (in order of arrival): 14 2 7 21 8 24
FCFS
SCAN
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
C–SCANCircular-SCAN moves head from edge to edge
serves requests on one way – uni-directionalimproves response time (fairness)
time
cylinder number1 5 10 15 20 25
12
incoming requests (in order of arrival):
14 2 7 21 8 24
schedulingqueue
24821721412
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
SCAN vs. C–SCANWhy is C-SCAN in average better in reality than SCAN when both service the same number of requests in two passes?
modern disks must accelerate (speed up and down) when seekinghead movement formula:
requests: navg. dist: xtotal cost:
requests: navg. dist: 2xtotal cost:
uni-directionalbi-directional
C-SCANSCAN
cylinders traveled
time
nβα + number of tracksseek time constantfixed overhead
xnxn ××=× 22 xnnxnxn ×+=×+× )(
22 22
2
nnnnn
nnn
++>×
+>cif n is large:
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
LOOK and C–LOOKLOOK (C-LOOK) is a variation of SCAN (C-SCAN):
same schedule as SCANdoes not run to the edgesstops and returns at outer- and innermost requestincreased efficiency SCAN vs. LOOK example:
time
cylinder number1 5 10 15 20 25
12
incoming requests (in order of arrival):
14 2 7 21 8 24
schedulingqueue
248
21
7
2
14
12
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
V–SCAN(R)V-SCAN(R) combines SCAN (or LOOK) and SSTF
define an R-sized unidirectional SCAN window, i.e., C-SCAN, and use SSTF outside the window
Example: V-SCAN(0.6) makes a C-SCAN window over 60 % of the cylindersuses SSTF for requests outside the window
V-SCAN(0.0) equivalent with SSTFV-SCAN(1.0) equivalent with C-SCAN
V-SCAN(0.2) is supposed to be an appropriate configuration
cylinder number1 5 10 15 20 25
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
What About Continuous Media?
Suitability of classical algorithmsminimal disk arm movement (short seek times)no provision of time or deadlinesgenerally not suitable
Continuous media server requirementsserve both periodic and aperiodic requestsnever miss deadline due to aperiodic requestsaperiodic requests must not starvesupport multiple streamsbalance buffer space and efficiency tradeoff
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Real–Time Disk Scheduling
Traditional algorithms have no provision of time or deadlines
Real-time algorithms targeted for real-time applications with deadlines
Several proposed algorithmsearliest deadline first (EDF)SCAN-EDFshortest seek and earliest deadline by ordering/value (SSEDO / SSEDV)priority SCAN (PSCAN)...
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Earliest Deadline First (EDF)EDF serves the request with nearest deadline first
non-preemptive (i.e., an arriving request with a shorter deadline must wait)
excessive seeks poor throughput
time
cylinder number1 5 10 15 20 25
12,5
incoming requests (<block, deadline>, in order of arrival):
14,6 2,4 7,7 21,1 8,2 24,3
schedulingqueue
12,5 14,6 2,4 7,7 21,1 8,2 24,3
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
SCAN–EDFSCAN-EDF combines SCAN and EDF:
the real-time aspects of EDFseek optimizations of SCANespecially useful if the end of the period of a batch is the deadline
increase efficiency by modifying the deadlinesalgorithm:
serve requests with earlier deadline first (EDF)sort requests with same deadline after track location (SCAN)
time
cylinder number1 5 10 15 20 25
2,3
incoming requests (<block, deadline>, in order of arrival):
14,1 9,3 7,2 21,1 8,2 24,2
schedulingqueue
2,3 14,1 9,3 7,2 21,1 8,2 24,2 16,116,1
Note:similarly, we can combine EDF with C-SCAN, LOOK or C-LOOK
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Stream Oriented Disk SchedulingStreams often have soft deadlines and tolerate some slack due to buffering, i.e., pure real-time scheduling is inefficient
Stream oriented algorithms targeted for streaming continuous media data
Several algorithms proposed:group sweep scheduling (GSS)mixed disk scheduling strategycontiguous media file system (CMFS)lottery schedulingstride schedulingbatched SCAN (BSCAN)greedy-but-safe EDF (GS_EDF)bubble up…
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Group Sweep Scheduling (GSS)GSS combines Round-Robin (RR) and SCAN
requests are serviced in rounds (cycles)principle:
divide S active streams into G groupsservice the G groups in RR orderservice each stream in a group in C-SCAN order playout can start at the end of the group
special cases:G = S: RR schedulingG = 1: SCAN scheduling
tradeoff between buffer space and disk arm movementtry different values for G giving minimum buffer requirement – select minimuma large G smaller groups, more arm movements, smaller buffersa small G larger groups, less arm movements, larger buffers
with high loads and equal playout rates, GSS and SCAN often service streams in same orderreplacing RR with FIFO and group requests after deadline gives SCAN-EDF
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Group Sweep Scheduling (GSS)GSS example: streams A, B, C and D g1:{A,C} and g2:{B,D}
RR group scheduleC-SCAN block schedule within a group
time
cylinder number1 5 10 15 20 25
A2 A1A3 B2 B3B1C1 C2 C3D3 D1 D2
g1
A2
C1
A1
A3
B2
B3
B1
C2
C3
D3
D1
D2
g2
g1
g2
g1
g2
{A,C}
{B,D}
{C,A}
{B,D}
{A,C}
{B,D}
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Mixed Disk Scheduling Strategy (MDSS)MDSS combines SSTF with buffer overflow and underflow prevention
data delivered to several buffers (one per stream)share of disk bandwidth allocated according to buffer fill levelSSTF is used to schedule the requests
......
share allocator
SSTF
sch
edul
er
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Continuous Media File System Disk Scheduling
CMFS provides (proposes) several algorithmsdetermines new schedule on completion of each requestorders requests so that no deadline violations occurdelays new streams until it is safe to proceed (admission control)
based on slack-time –based on amount of data in buffers and deadlines of next requests(how long can I delay the request before violating the deadline/buffer running empty?)
defines the amount of time that o can be used for non-real-time requests oro work-ahead for continuous media requests
useful algorithmsgreedy – serve one stream as long as possiblecyclic – distribute current slack time to maximize future slack timeboth always serve the stream with shortest slack-time, but cyclic is more CPU intensive
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Mixed Media Oriented Disk Scheduling
Multimedia applications may require both RT and NRT data – desirable to have all on same diskTargeted for various streaming continuous media data
Several algorithms proposed:Felini’s disk schedulerDelta LFair mixed-media scheduling (FAMISH)MARS schedulerCello Adaptive disk scheduler for mixed media workloads (APEX) …
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
MARS Disk SchedulerMassively-parallel And Real-time Storage (MARS) scheduler supports mixed media on a single system
a two-level schedulinground-based
top-level:1 NRT queue and n (1) RT queue(SCAN, but future GSS, SCAN-EDF, or…)
use deficit RR fair queuing to assign quantums to each queue per round –divides total bandwidth among queues
bottom-level:select requests from queues according to quantums, use SCAN order
work-conserving(variable round times, new round starts immediately)
…
deficit round robin fair queuingjob selector
NRT RT
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
CelloCello is part of the Symphony FS supporting mixed media
two-level schedulinground-based
top-level: n (3) service classes (queues)deadline (= end-of-round) real-time (EDF)throughput intensive best effort (FCFS)interactive best effort (FCFS)
divides total bandwidth among queues according to a static proportional allocation scheme(equal to MARS’ job selector)
bottom-level: class independent scheduler (FCFS)select requests from queues according to BW sharesort requests from each queue in SCAN order when transferred
partially work-conserving(extra requests might be added at the end of the classindependent scheduler if space, but constant rounds)
deadline RT throughput intensivebest-effort
interactivebest-effort
31
27
84
2 1 2
sort each queue in SCAN order when transferred
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Adaptive Disk Scheduler for Mixed Media Workloads
APEX is another mixed media schedulertwo-level, round-based scheduler similar to Cello and MARS
uses token bucket for traffic shaping(bandwidth allocation)
the batch builder select requests inFCFS order from the queues based on number of tokens – each queue must sort according to deadline (or another strategy)
work-conservingadds extra requests if possible to a batchstarts extra batch between ordinary batches
Request Distributor/Queue Scheduler
Queue/BandwidthManager
...
Batch Builder
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
APEX, Cello and C–LOOK ComparisonResults from Ketil Lund (2002)Configuration:
Atlas Quantum 10KAvg. seek: 5.0msAvg. latency: 3.0mstransfer rate: 18 – 26 MB/s
data placement: random, video and audio multiplexedround time: 1 secondblock size: 64KB
Video playback and user queriesSix video clients:
each playing back a random videorandom start time (after 17 secs, all have started)
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
APEX, Cello and C–LOOK ComparisonNine different user-query traces, each with the following characteristics:
Inter-arrival time of queries and disk requests are exponentially distributed
Start with one trace, and then add traces, in order to increase workload (⇒ queries may overlap)
Video data disk requests are assigned to a real-time queueUser-query disk requests to a best-effort queue
Bandwidth is shared 50/50 between real-time queue and best-effort queue
We measure response times (i.e., time from request arrived at disk scheduler, until data is placed in the buffer) for user-query disk requests, and check whether deadline violations occur for video data disk requests
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
APEX, Chello and C–LOOK ComparisonAverage response time for user-query disk requests
0
200
400
600
800
1000
1200
1400
1600
1800
2000
1 2 3 4 5 6 7 8 9
# User-query traces
Res
pons
e tim
e (m
s)
APEXCelloC-LOOK
32662059127181140428890180C-LOOK
000000000Cello
000000000APEX
987654321
Deadlineviolations(video)
Data Placement on Disk
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Data Placement on Disk
Disk blocks can be assigned to files many ways, and several schemes are designed for
optimized latencyincreased throughput
access pattern dependent
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Layout
Constant angular velocity (CAV) disksequal amount of data in each track(and thus constant transfer time)constant rotation speed
Zoned CAV diskszones are ranges of trackstypical few zonesthe different zones have
different amount of data different bandwidth i.e., more better on outer tracks
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk Layoutouter:
inner:
outer:
inner:
constant transfer rate
variabletransfer rate
zoned disk
non-zoned disk
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Disk LayoutCheetah X15.3 is a zoned CAV disk:
4192,69075,3%8188848596,07438234210
4428,26875,5%8648960632,4746623259
4781,50675,7%9338880649,4148024378
5345,64176,3%10440704687,0551225547
5875,00375,5%11474616728,4753726766
6603,66978,1%12897792755,2957628055
7148,07376,0%13961080801,8859529394
7854,29376,5%15340416835,7662430793
9013,24876,0%17604000878,4365233822
9735,63577,2%19014912890,9867235441
Formatted Capacity (MB)Efficiency
Sectors per Zone
Zone Transfer Rate (MBps)
Sectors per Track
Cylinders per ZoneZone
Always place often used or high rate data on outermost tracks (zone 1) …!?
NO, arm movement is often more important than transfer time
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Data Placement on DiskContiguous placement stores disk blocks contiguously on disk
minimal disk arm movement reading the whole file (no intra-file seeks)
pros/cons☺ head must not move between read operations - no seeks / rotational delays☺ can approach theoretical transfer rate
but usually we read other files as well (giving possible large inter-file seeks)
real advantagedo not have to pre-determine block (read operation) size (whatever amount to read, at most track-to-track seeks are performed)
no inter-operation gain if we have unpredictable disk accesses
file A file B file C
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Using Adjacent Sectors, Cylinders and Tracks
To avoid seek time (and possibly rotational delay), we canstore data likely to be accessed together on
adjacent sectors (similar to using larger blocks)
if the track is full, use another track on the same cylinder (only use another head)
if the cylinder is full, use next (adjacent) cylinder (track-to-track seek)
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Data Placement on DiskInterleaved placement tries to store blocks from a file with a fixed number of other blocks in-between each block
minimal disk arm movement reading the files A, B and C(starting at the same time)
fine for predictable workloads reading multiple files
no gain if we have unpredictable disk accesses
Non-interleaved (or even random) placement can be used for highly unpredictable workloads
file Afile B
file C
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Data Placement on DiskOrgan-pipe placement consider the ‘average’ disk head position
place most popular data where head is most often
center of the disk is closest to the head using CAV disksbut, a bit outward for zoned CAV disks (modified organ-pipe)
disk:innermost
outermost
head
bloc
k ac
cess
pro
babi
lity
cylinder number
bloc
k ac
cess
pro
babi
lity
cylinder number
organ-pipe: modified organ-pipe:Note:skew dependent on
tradeoff between zoned transfer time and storagecapacity vs.
seek time
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
BSD Example: Fast File System
FFS is a general file systemidea is to keep inode and associated blocks close(no long seeks when getting the inode and data)
organizes the disks in partitions – cylinder groupshaving several inodesfree block bitmap…
tries to store a file within a cylinder groupnext block on same cylindera block within the cylinder groupfind a block in another group using a hash functionsearch all cylinder groups for a free block
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
BSD Example: Log-Structured File SystemLog-structured placement is based on assumptions (facts?) that
RAM memory is getting largerwrite operations are most expensivereads can often be served from buffer cache (!!??)
Organize disk blocks as a circular log
periodically, all pending (so far buffered) writes are performed as a batchwrite on next free block regardless of content (inode, directory, data, …)
a cleaner reorganizes holes and deleted blocks in the background
stores blocks contiguously when writing a single fileefficient for small writes, other operations as traditional UNIX FS
disk:
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Linux Example: XFS, JFS, …Count-augmented address indexing in the extent sections
observation: indirect block reads introduce disk I/O and break access locality
introduce a new inode structure
add counter field to data pointer entries – direct points to a disk blockand count indicated how many other blocks is following the first block (contiguously)
if contiguous allocation is assured, each direct entry is able to access much more blocks without additional retrieving an indirect block
direct 0
direct 1
direct 2
…
direct 10
direct 11
triple indirect
single indirect
double indirect
attributes
count 0
count 1
count 2
…
count 10
count 11
data3 data data
inode
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Windows Example: NTFSEach partition contains a master file table (MFT)
a linear sequence of recordseach record describes a directory or a file (attributes and disk addresses)
first 16 reserved forNTFS metadata
record header
standard info
file name
data header
info about data blocks
…data…
A file can be …
• stored within the record (immediate file, < few 100 B)
• represented by disk block addresses (which hold data):runs of consecutive blocks (<addr, no>, like extents)
• using several records if more runs are needed
20 4
run 1
30 2
run 2
74 7
run 3unused
24 - base record
26 - first extension record
27 - second extension record
10 2
run 1
78 3
run k
MFT 27
2nd extension
MFT 26
1st extension
run 2, run 3, …, run k-1
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
BSD Example: Minorca File System Minorca is a multimedia file system (from IFI/UiO)
enhanced allocation of disk blocks for contiguous storage of media filessupports both continuous and non-continuous files in the same system using different placement policies
Multimedia-Oriented Split Allocation (MOSA) –one file system, two sections:
cylinder group sections for non-continuous fileslike traditional BSD FFS disk partitionssmall block sizes (like 4 or 8 KB)traditional FFS operations
extent sections for continuous filesextents contain one or more (adjacent) cylinder group sections
o summary informationo allocation bitmapo data block area
expected to store one media filelarge block sizes (e.g., 64 KB) new “transparent” file operations, create file using O_CREATEXT
cylinder group
cylinder group
…
extent
extent
extent
…
extent
extent
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Other File Systems Examples
Contiguous AllocationPresto
similar to Minorca extents for continuous filesdoesn’t support small, discrete files
Fellinisimple flat file systemmaintains free block list with grouping contiguous blocks
Continuous Media File System...
Modern Disks:Complicating Factors
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Complicating FactorsDisk used to be simple devices and disk scheduling used to be performed by OS (file system or device driver) only…
… but, new disks are more complex hide their true layout, e.g.,
only logical block numbersdifferent number of surfaces, cylinders, sectors, etc.
OS view real view
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Complicating FactorsDisk used to be simple devices and disk scheduling used to be performed by OS (file system or device driver) only…
… but, new disks are more complex hide their true layouttransparently move blocks to spare cylinders
Seagate X15.3 - zone 1 - 10: (7,7,6,6,6,5,5,5,5,5)e.g., due to bad disk blocks
OS view real view
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Complicating FactorsDisk used to be simple devices and disk scheduling used to be performed by OS (file system or device driver) only…
… but, new disks are more complex hide their true layouttransparently move blocks to spare cylindershave different zones
OS view real view
NB! illustration of transfer time, not rotation speed
Constant angular velocity (CAV) disks
constant rotation speedequal amount of data in each trackthus, constant transfer time
Zoned CAV disksconstant rotation speed zones are ranges of trackstypical few zonesthe different zones have different amount of data, i.e., more better on outer tracksthus, variable transfer time
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Complicating FactorsDisk used to be simple devices and disk scheduling used to be performed by OS (file system or device driver) only…
… but, new disks are more complex hide their true layouttransparently move blocks to spare cylindershave different zoneshead accelerates – most algorithms assume linear movement overhead
~ 10x - 20x
x
1 NCylinders Traveled
Time
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Complicating FactorsDisk used to be simple devices and disk scheduling used to be performed by OS (file system or device driver) only…
… but, new disks are more complex hide their true layouttransparently move blocks to spare cylindershave different zoneshead accelerateson device (and controller) buffer caches may use read-ahead prefetching
diskbufferdisk
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Complicating FactorsDisk used to be simple devices and disk scheduling used to be performed by OS (file system or device driver) only…
… but, new disks are more complex hide their true layouttransparently move blocks to spare cylindershave different zoneshead accelerateson device (and controller) buffer caches may use read-ahead prefetchinggaps and checksums between each sector
track
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Complicating FactorsDisk used to be simple devices and disk scheduling used to be performed by OS (file system or device driver) only…
… but, new disks are more complex hide their true layouttransparently move blocks to spare cylindershave different zoneshead accelerates on device (and controller) buffer caches may use read-ahead prefetchinggaps and checksums between each sector
“smart” with a build-in low-level scheduler (usually SCAN-derivate)we cannot fully control the device (black box)
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Why Still Do Disk Related Research?
If the disk is more or less a black box – why bother?
many (old) existing disks do not have the “new properties”
according to Seagate technical support:
“blocks assumed contiguous by the OS probably still will be contiguous, but the whole section of blocks might be elsewhere”
[private email from Seagate support]
delay sensitive requests
But, the new disk properties should be taken into accountexisting extent based placement is probably goodOS could (should?) focus on high level scheduling only
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Next Generation Disk Scheduling?Thus, due to the complicating factors…(like head acceleration, disk buffer caches, hidden data layout, built-in “SCAN” scheduler,…)
… an VoD server scheduler can be (???):
hierarchical high-level software schedulerseveral top-level queues (at least RT & NRT)
process queues in rounds (RR)o dynamic assignment of quantumso work-conservation with variable round length
(full disk bandwidth utilization vs. buffer requirement)
only simple collection of requests according to quantums in lowest level and forwarding to disk, because ...
...fixed SCAN scheduler in hardware (on disk)
Current research also look at on-device programmable processors
…
RT NRT
SCAN
EDF / FCFS
No sorting
The End:Summary
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
SummaryThe main bottleneck is disk I/O performance due to disk mechanics: seek time and rotational delays
Many algorithms trying to minimize seek overhead(most existing systems uses a SCAN derivate)
Proper block allocation may improve transfers rate
World today more complicated (both different media and unknown disk characteristics)
Next week, storage systems (part II)multiple disksmultimedia file system examples...
2005 Carsten Griwodz & Pål HalvorsenINF5070 – media servers and distribution systems
Some References1. Anderson, D. P., Osawa, Y., Govindan, R.:”A File System for Continuous Media”, ACM Transactions on
Computer Systems, Vol. 10, No. 4, Nov. 1992, pp. 311 - 3372. Elmasri, R. A., Navathe, S.: “Fundamentals of Database Systems”, Addison Wesley, 20003. Garcia-Molina, H., Ullman, J. D., Widom, J.: “Database Systems – The Complete Book”, Prentice Hall, 20024. Lund, K.: “Adaptive Disk Scheduling for Multimedia Database Systems”, PhD thesis, IFI/UniK, UiO 5. Plagemann, T., Goebel, V., Halvorsen, P., Anshus, O.: “Operating System Support for Multimedia Systems”,
Computer Communications, Vol. 23, No. 3, February 2000, pp. 267-289 6. Seagate Technology, http://www.seagate.com7. Sitaram, D., Dan, A.: “Multimedia Servers – Applications, Environments, and Design”, Morgan Kaufmann
Publishers, 2000