Capacity Architecture .


22 views
Uploaded on:
Description
Storage Architecture. CE202 December 2, 2003 David Pease. Faster. Smaller. Higher. Cache. RAM. Capacity. Speed. Cost. Disk. Optical. Tape. Slower. Larger. Lower. Hierarchy of Storage. Storage System Components. Application I/O Library File System Device Driver
Transcripts
Slide 1

Capacity Architecture CE202 December 2, 2003 David Pease

Slide 2

Faster Smaller Higher Cache RAM Capacity Speed Cost Disk Optical Tape Slower Larger Lower Hierarchy of Storage

Slide 3

Storage System Components Application I/O Library File System Device Driver Host Bus Adapter Interconnect Storage Controller Devices I/O Context

Slide 4

Disks

Slide 5

Disk Drives "Workhorse" of advanced stockpiling frameworks Capacity expanding, crude value dropping can purchase 1TB for just $1000! transmission capacity not keeping pace unwavering quality is really diminishing monstrous frameworks can mean even lower accessibility Majority of cost of proprietorship in organization, not price tag reinforcement, design, disappointment recuperation

Slide 6

Disk Architecture shaft barrel part track platters arms with read/compose heads revolution

Slide 7

Disk Storage Density

Slide 8

Disk Capacity Growth

Slide 9

IBM Disk Storage Roadmap

Slide 10

Storage Costs

Slide 11

RAID Redundant Arrays of Inexpensive Disks Two orthogonal ideas: information striping for execution excess for dependability Striped exhibits can build execution, yet at the cost of dependability (next page) repetition can give clusters preferred unwavering quality over an individual circle

Slide 12

Reliability of Striped Array

Slide 13

One-month Trace of Hardware Failures Trace gathered from the Internet Archive (March 2003) (expresses gratitude toward Kelly Gottlib) - Over 100 terabytes of packed information - 30 circle disappointments out of aggregate 70 equipment issues

Slide 14

RAID Levels

Slide 15

RAID Levels 0 1 2 3 4 5 6

Slide 16

RAID: 4x Small Write Penalty little information compose xor 3 4 1 2 5

Slide 17

Log-Structured File Systems Based on suspicion that circle movement will get to be commanded by composes Always composes plate information consecutively, into next accessible area on plate no looks for on compose Eliminates issue of 4x compose punishment all composes are "new", no compelling reason to peruse old information or equality However, no cases in industry record frameworks

Slide 18

Tape

Slide 19

Tape Media Inherently successive long time to first byte no irregular I/O Subject to mechanical push number of read-compose cycles lower than circle Problems as a documented medium: perusers leave after a few years most quickly lately tapes (with information) stay in a salt mine

Slide 20

Tape Media Density will dependably trail that of circle Tape extends, more hard to get higher thickness Alignment additionally an issue once it\'s past the head, it\'s gone more traditionalist strategies required Bottom line: mechanical building issues for tape are the troublesome ones

Slide 21

Optical CD, CD-R/RW, DVD, DVD-R/RW Capacities: CD: ~700MB (gigantic 20 years prior!) DVD: single sided, single layer: 5GB single sided, twofold layer: 9GB twofold sided, single layer: 10GB twofold sided, twofold layer: 18GB Size of cell restricted by wavelength of light current lasers are red blue lasers are being worked on, then UV, ...

Slide 22

Optical Magneto-optical (HAMR) warm from laser makes altering course of charge less demanding (so cell is littler)

Slide 23

MEMS MicroElectroMechanical Systems 6-10 times quicker than plate cost and limit issues

Slide 24

Magnetic RAM (MRAM) Stores every piece in an attractive cell as opposed to a capacitor or flip-tumble information is industrious Can be perused and composed rapidly Read and compose times 0.5 – 10 µs or less Individual bits are writeable (no square eradicate) Density & cost tantamount to DRAM may require thickness/speed tradeoffs denser MRAM may need to run slower on account of warmth dissemination on composes

Slide 25

Magnetic RAM (MRAM) Several organizations have reported associations to deliver items ~2003 Ideas for utilization of MRAM away: Persistent reserve Hot information in MRAM, icy information to circle No compelling reason to flush compose reserve to maintain a strategic distance from information misfortune HeRMES all metadata in MRAM enough document information in MRAM to shroud plate inertness for first access to a record

Slide 26

Peripheral Busses SCSI IDE/ATA HIPPI (High Performance Parallel Intf.) IEEE 1394 (FireWire) FibreChannel (FCP) IP (e.g., iSCSI) InfiniBand Serial ATA

Slide 27

Peripheral Busses Parallel SCSI, most printers, IBM Channels at least 1 bytes for each time Skew issues at high speeds Serial FC, RS232, IEEE1394 (FireWire) 1 bit for each clock, self timing can be keep running at much higher paces than parallel transport

Slide 28

Networked Storage joined by universally useful or devoted system (e.g., FibreChannel) Motivations: homogenous and heterogeneous record sharing incorporated organization better asset use (shared capacity assets, pooling) Dedicated Networks: Fiber-Channel: FCP (SCSI over FC) iSCSI: SCSI over IP InfiniBand

Slide 29

Networked Storage Can mean numerous things: NAS (Network-Attached Storage): document server machines serving NFS as well as CIFS (for instance, Network Appliance) NASD (Network-Attached Secure Disk): smart, arrange appended drives w/security highlights (likewise, Network-Attached Storage Device) SAN (Storage Area Network): arrange for connecting circles and PCs, normally committed just to capacity operations OBSD (Object-Based Storage Device): like NASD

Slide 30

Solaris Win2K Linux AIX IFS w/store IFS w/store IFS w/reserve IFS w/store A SAN File System NFS CIFS FTP HTTP Control Network (IP) Meta-information Server Meta-information Server Meta-information SAN Meta-information Server information Security helps Storage Management Server HSM & Backup Data

Slide 31

Additional Reading Hennessy & Patterson: Chapter 6 Chen, Lee, Gibson, Katz, & Patterson: RAID: elite, dependable optional stockpiling. ACM Computing Surveys 26, June 1994, 145-185 Rosenblum & Ousterhout: The outline and usage of a log-organized document framework. ACM Transactions on Computer Systems, Feb. 1992, 26-52 Gibson, Nagle, et al.: A savvy, high-transfer speed stockpiling design. Procedures of the Eight Conference on Architectural Support for Programming Languages and Operating Systems, 1998 http://www.almaden.ibm.com/cs/storagesystems/stortank/

Recommended
View more...