ECE6160: Advanced Computer Networks

1
ECE6160 Advanced Computer Networks

• iSCSI
• Xubin (Ben) He

2
The Evolution of Storage Architectures
3
(No Transcript)
4
The Convergence of Networking and Storage
Host
I/O Application
File System
Networking TCP, UDP, IP GBE,
10GBE
Storage Parallel SCSI Fibre
Channel IDE, ATA
TCP/IP Network
iSCSI
Storage
iSCSI-based Storage Area Network (SAN)(block
level)
5
What is iSCSI

• The protocol for transfer of SCSI commands and
  data over TCP/IP networks is called iSCSI.

6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
iSCSI Session

• iSCSI works in the session context
• A session comprises one or more TCP connections
  between the initiator and target
• A session has four phases
• Initial login phase
• Security Authentication Phase (optional)
• Operational Negotiation Phase (optional)
• Full Featured Phase
• The first three phases normally occur at the
  system start time (in iSCSI driver).
• iSCSI data and command exchange in the
  full-featured phase.

12
A Normal iSCSI Setup Process

• Discovery Session
• Goal find all available targets
• Establish TCP connection with the gateway
• Initial login between the initiator and gateway
• Parameter negotiation (security phase omitted)
• Obtain all available targets
• Setup Session
• Goal initiator establish a session with each
  target
• Establish TCP connection for each target
• Initial Login with each target
• Parameter negotiation (security phase omitted)
• Full featured phase

13
iSCSI Target Identification

• Static Configuration
• The initiator statically configure the IP address
  or target name
• Used in a small number of initiators
• Service Location Protocol (SLP)
• Use SLP protocol to discover targets
• Adequate for medium sized network
• Internet Service Naming Service (iSNS)
• Use iSNS protocol to discover targets
• Internet

14
References

• iSCSI RFC (IETF), http//www.faqs.org/rfcs/rfc3720
  .html, April 2004.
• HP Tech. Report, SCSI over TCP/IP, Sept. 2000
• IBM Research Lab, iSCSI-protocol and
  implementation, Sept. 2000
• Cisco Tech. Report, Jan. 2001

15
Performance Evaluation of Distributed iSCSI RAID

• International workshop on Storage Network
  Architecture and Parallel I/Os (SNAPI'03), held
  with
• 12th IEEE/ACM International Conference on
  Parallel Architectures and Compilation Techniques
  (PACT)
• Xubin He, Praveen, Dan
• Storage Technology Architecture Research(STAR)
  Lab
• Department of Electrical and Computer Engineering
• Tenessee Technological University, U.S.A

16
Outline

• Introduction
• Architecture and Design
• Evaluations
• Conclusions and Future Work

17
Motivations

• iSCSI is emerging as a new protocol to deploy
  storage over IP using Ethernet.
• Compatability existing SW/HW
• Economic
• But???
• Performance
• Reliability

18
Existing work

• Protocol Development
• Satran et al, Features of iSCSI Protocol, IEEE
  Communications.
• Performance Evaluations/Improvements
• X. He et al, iCache, IEEE LCN2002
• Grunwald et al, A Performance Analysis of iSCSI
  Protocol, IEEE Mass Storage, 2003
• David Du et al, Performance Study of iSCSI-base
  Storage Subsystems, IEEE Communications, 2003
• Implementation
• Meth, iSCSI initiator design and implementation
  experience, IEEE Mass Storage, 2002
• IBM,CISCO,UNH,UMass, URI

19
RAID

• Striping to improve performance
• Redundancy (parity) to improve reliability
• Local
• Distributed RAID special interconnection/control
  ler
• TickerTAIP

• Can we apply RAID technique to iSCSI networked
  storage using Ethernet?

20
iRAID iSCSI RAID

• Striping data across different iSCSI targets
• Use one target as redundancy to improve
  reliability
• Traditional RAID
• a disk/partition is a storage unit
• RAID controller
• iRAID
• an iSCSI target is a unit
• Ethernet switch as the controller

21

• Introduction
• Architecture and Design
• Evaluations
• Conclusions and Future Work

22
iRAID
23
S-iRAID Striping data across multiple targets
Ethernet Switch
24
P-iRAID iRAID with parity target
Ethernet Switch
iSCSI Targets
25

• Introduction
• Architecture and Design
• Evaluaitons
• Conclusions and Future Work

26
Experimental Setup
27
Experimental setup

• Chunk size (block size) 64KB
• Benchmark Iozone
• Iozone ra s dataset size r request size P i0
  i1 I2 f /mnt/iRAID/test

28
Throughput(1GB, Gbps, 64KB)
S-iRAID 6.6, and P-iRAID 2.17
29
Where is the bottleneck?

• Initiator?
• Network Bandwidth?
• Target?

30
Different BW
For low-speed network, BW is the bottleneck.
31
Different networks/targets
With increasing BW, iSCSI targets may become the
bottleneck.
32
Different of targets
33
iCache
34
iRAID vs Hardware RAID
iRAID 4 targets, 256MB RAM each, 500 x4
2000 Dell server w/ PERC RAID controller 1GB
RAM, 4x seagate SCSI, 3000
35
Reliability

• Mounted 4-target P-iRAID as a local drive
  (/mnt/p-iRAID) on the initiator
• Copied the Linux source tree (/usr/src) to this
  P-iRAID drive
• Rebooted all those machines (initiator and
  targets), and formatted one of the iSCSI target
  drive (this will erase all data on it) to emulate
  one target failure
• Mounted the 4-target P-iRAID as a local drive
  (/mnt/p-iRAID) on the initiator again
• Compiled the Linux source tree /mnt/p-iRAID/usr/sr
  c successfully.

36
Conclusions

• Both S-iRAID and P-iRAID improve performance
• P-iRAID improves the reliability

37
References

• Satran et al, Features of iSCSI Protocol, IEEE
  Communications.
• X.He, Praveen, Dan, iRAID, IEEE/ACM SNAPI2003
• X. He et al, iCache, IEEE LCN2002
• Grunwald et al, A Performance Analysis of iSCSI
  Protocol, IEEE Mass Storage, 2003
• David Du et al, Performance Study of iSCSI-base
  Storage Subsystems, IEEE Communications, 2003
• Meth, iSCSI initiator design and implementation
  experience, IEEE Mass Storage, 2002