Transcript MS PowerPoint - Southern Illinois University Edwardsville

Department of Computer Science
Southern Illinois University Edwardsville
Dr. Hiroshi Fujinoki and Kiran Gollamudi
E-mail: {hfujino, kgollam}@siue.edu
Problem Definition
Response Time
The elapsed time between the end of an inquiry on
a computer system and the beginning of a response
Problem
• Long response time
• Transmission time
• Operating system overhead
Due to high web traffic load
FAT look-up, following a sector-chain,
multiple clients etc
Delay Causes
We have 3 different causes of delay
Operating system overhead
RECEIVER
SENDER
Internet
Internet
Routing Overhead + Error/Flow control Overhead
Network Protocol Processing Overhead
HTTP Client/Server model
Terms Defined
HTTP Client
HTTP Server
TCP Sync
TCP Sync ACK
• Response time
HTTP Get
• Transmission time
Transmittingrequested file
Requested file
starts arriving.
Time
Existing work for Client/Server model
The following are the existing techniques to reduce response time
and transmission delay
Existing Methods
Caching
Mirroring
Server Clustering
Object Packaging
Objective
To improve response time and transmission time by :
• Single request for the multiple files
• Reduced network protocol overhead during the transmission at routers
 CRC calculation overhead
 Memory copies
• Reduced number of packets by minimizing fragmentation
• Reduced OS overhead at a web server
 Reduced FAT lookups
 Flow control and error control
Concept Of Object Packaging
Object Packaging:
• Is a collection of web files in a web site
• Files are sequentially packed without compression
Number of objects
Pointer
File #1
Object Offset Fields
File header
Object Package
File #n
Packaging Format Of Object Packaging
Object Information Field
File name subfield
sub
File size subfield
sub
File attribute subfield
sub
Number of objects (files) subfield
sub
• Object Information Field: Contains the information of the packed files
• Data Field: Contains the contents of the requested files
Sub field
•A
collection
of the
the sizes
names
of
the
files
Number
of objects:
Tells
how
many
packed
filesorare
there
• •File
Attribute
Subfield:
Each
file
is contained
a binary,
text,
executable
•A
collection
of
of
the
contained
files
Experiment Modeling
Test-Bed
Requested Files
Requests
Hub
Client
Web Server
Local Disk
Traffic Monitor
• Client Machine: The host that requests the server for the data
• Hub : Broadcasts every packet to every port
• Traffic Monitor: Monitors the all the traffic (packets) that are passing
Experiment Design
Experimental Setup
• Files with 1K, 4K and 10K bytes are placed in the server machine
• Factors measured for both the existing method and
object Packaging
1. Average response time
2. Average bytes transferred
3. Number of Transferred packets
4. Transmission time
Experimental Output
• The figures represent all factors measured with different file sizes.
File 4K
Existing Method
Percentage
100%
Object Packaging
80%
60%
40%
20%
0%
Avg. Time
No.Of Pkts
Bytes transferred
File 10K Time
Existing Method
100%
Object Packaging
Percentage
80%
60%
40%
20%
0%
Avg. Time
No.Of Pkts
Bytes transferred
•Percentage to the existing method
Experimental Output
•100%
•Transferred bytes
•80%
•Response time
•60%
•40%
•Transferred packets
•20%
•0%
•1K
•4K
•10K
•Average file size (in bytes)
File size = 4K
Percentage to the existing method
100%
File size = 10K
80%
File size = 1K
60%
Number of transferred
packets relative to the
existing method
40%
20%
0%
0
25
50
Number of transferred files
75
100
Summary
• Multiple file transmissions by object packaging proposed
• No modification of an operating system at the server side nor
transmission protocol at routers required
• Object packaging is efficient in reducing response time and
transmission load and time
Future work
There are two on-going activities and future works
On Going Activities
1. Measuring the CPU load, response time and propagation
delay on the server
2. Measuring all the above but for multiple clients
Future Works
1. Perform the experiments using a Gigabit Ethernet cable
2. To observe the scalability
References
1. M. Arlitt and C. Williamson, “Web Server Workload Characterization: The
Search for Invariants,” Proceedings of the 1996 ACM SIGMETRICS
Conference on the Measurement and Modeling of Computer Systems,”
May 1996, pp. 126-137
2. GVU’s WWW User Surveys, Georgia Institute of Technology
URL: http://www.gvu.gatech.edu/user_surveys
3. J. Ousterhout, “Why Aren't Operating Systems Getting Faster As
Hardware?,” Proceedings of Summer 1990 USENIX Conference,
June 1990, pp. 247-256
4. P. Druschel, “Operating System Support for High-Speed Networking,”
Communications of the ACM, vol. 39, no. 2, September 1996, pp. 41-51
5. P. Markatos, “Speeding-up TCP/IP: Faster Processors Are not Enough,”
Proceedings of the 21st IEEE International Performance, Computing,
and Communications Conference, April 2002, pp. 341-345
References
6. M. Busari and C. Williamson, “On the Sensitivity of Web Proxy Cache
Performance to Workload Characteristics,” Proceedings of IEEE
INFOCOM,April 2001, pp. 1225-1234
7. J. Dilley, “The Effect of Consistency on Cache Response Time,” IEEE
Network, vol. 14, no. 3, May/June 2000, pp. 24-28
8. S. Glassman, “A caching relay for the Worldwide Web,” Computer Networks and ISDN Systems, vol. 27, no. 2, October 1994, pp. 165-173
9. D. Lee, “PreFetch Document Caching to Improve Worldwide Web User
Response Time,” Master's Thesis. Virginia Polytechnic Institute and State
University, March 1996
10. J. Mogul, “Squeezing More Bits Out of HTTP Caches,” IEEE Network,
vol. 14, no.3, May/June 2000, pp. 6-14
11. Figures of Red hat and Windows from the internet sites.
The End
Thank you !!!
Client Side Caching
Advantages
• Reduced latency
• Reduced server load
• Reduced bandwidth consumption in a network
Disadvantages
• Effective only when same files are repeatedly requested
• Additional hardware or expertise is required
• No benefit if object is not cached
• May be unable to cache multimedia content
• Unable to cache dynamically generated content
Go To Previous Slide
Mirroring Technique
Advantages
• Improves throughput
• Complete redundancy of data
• Fast recovery from a disk failure
Disadvantages
• Expensive
• Low capacity
• No improvement in data access speed
Go To Previous Slide
Server Clustering
Advantages
• Fail over
• Scalability
• Fault resilience
• Load balancing
Disadvantages
• Requires investment for hardware
• Request dispatcher may be a bottleneck
Go To Previous Slide