0% found this document useful (0 votes)
58 views9 pages

NPTEL Course Jan 2012: Storage Systems

This document discusses storage systems and their characteristics. It covers various abstractions in the storage stack including devices, blocks, files and applications. It describes how storage systems have become more distributed with separation of processing and storage. The document also discusses performance optimization techniques for storage like caching, parallelism across devices, and workload analysis. Storage protocols use interrupt-driven techniques and split-phase transactions to handle slow storage speeds.

Uploaded by

Souptik Sen
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
58 views9 pages

NPTEL Course Jan 2012: Storage Systems

This document discusses storage systems and their characteristics. It covers various abstractions in the storage stack including devices, blocks, files and applications. It describes how storage systems have become more distributed with separation of processing and storage. The document also discusses performance optimization techniques for storage like caching, parallelism across devices, and workload analysis. Storage protocols use interrupt-driven techniques and split-phase transactions to handle slow storage speeds.

Uploaded by

Souptik Sen
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Storage Systems

NPTEL Course
Jan 2012
(Lecture 04)
K. Gopinath
Indian Institute o Science

!"t2 #S
(rom $i%ipedia)


Goog&e #S
(rom $i%ipedia)


'eep Storage Stac%

(arious types o a)stractions in stac%

'e*ice

+&oc%

#i&e

,pp&ication &e*e& ()uering in &i)c- or eg.)

#iner su)&ayers in each &ayer

SCSI has upper (de*ice.speciic)- mid (protoco& speciic)


and &o/er (physica& communication &ayer)

#or sca&a)i&ity- net/or% stac% part o storage stac%

, good part o stac% in %erne&

Increasing&y- storage stac% migrating out o %erne& /ith


separation o processing and storage (eg. G#S)

Storage Characteristics

Concurrency arises natura&&y

$ide disparity in speeds o memory and storage

0a*e to mas% s&o/ness o storage

1a%e processing and storage go at their o/n rates and


use interrupts (or sometimes po&&ing) to signa& comp&etion
o s&o/ storage operations

0istoric reason /hy operating systems de*e&oped

Storage has to )e typica&&y persistent o*er time

,mount increases typica&&y /ith time

+ut not a&& imp o*er time2 %eep imp part in ast storage3

4Caching5 and 4tiering5 arise natura&&y

0a*e to choose 2 o 67 speed- capacity or cost



Storage 8erormance

Storage oten s&o/est component

Cache9

$ithin sing&e de*ice- eiciency )y7

merging re:uests

schedu&ing re:uests in an order that is )est /rt de*ice


(out.o.order e"ecution commonp&ace)

0igher &e*e& sot/are has to /or% around this aspect

I a particu&ar order re:uired- &et to 4user5

Semantica&&y not much guaranteed

,synchronous processing oten used7 aio

8ara&&e&ism across mu&tip&e de*ices;threads7

1u&tip&e 0eads ('is%s)

1u&tip&e chips (SS's)



<ptimi=ation #rame/or%

'ue to s&o/ness o de*ices- optimi=ation o accesses


important

eg. /hat to cache- /hat to preetch3

+ut usage patterns typica&&y not %no/n a priori

+ig dierence in perormance /hether se:uentia& access


or random

System s&o/ i too many on demand migrations rom


s&o/ to ast tier o storage (&atency de&ays)

<ten- opts. critica& and o*erride 4semantics5

<ut.o.order processing typica&

Comp&e" higher.&e*e& sot/are

Learning on the >o) important

Simp&e and ro)ust methods useu&



Storage 8rotoco&s

Interrupt dri*en rather than /ait;po&&

<n comp&etion- interrupt C8? or 0+,

@o a*oid interrupt o*erhead- 0+, or simi&ar agents

0e&ps Segmentation and Aeassem)&y (S,A)

Sp&it.phase transactions common

or eg7 on comp&etion o (a &ong) see%- s&a*e ta%es )us

8rotoco& endpoints preera)&y 4*irtua&i=a)&e5

SCSI de*ices can )e on an e&ectrica& )us- net/or% or Internet


i physica& &ayer hand&ed correct&y

8rotoco&s sur*i*e much &onger

'e*ices can ha*e ar)itrary structure as &ong as they spea%


SCSI protoco&

!*en )ig ser*ers9



Summary

Storage systems design has many ramiications


or the rest o the system

8ro*ide a)stractions )ased on app&ication


needs and de*ices

'esign needs to )e sensiti*e to cost- de*ices-


managea)i&ity

Introduce ne/er a)stractions /ith time

eg. %ey *a&ue stores

Storage systems need to sca&e to support &arge


sca&e computing systems

You might also like