LCN2003001

1
Web File Transmission by Object
Packaging Performance Comparison with HTTP 1.0
and HTTP 1.1 Persistent Connection
Hiroshi Fujinoki, Murugesan Sanjay and Chintan
Shah
Department of Computer Science School of
Engineering Southern Illinois University
Edwardsville
LCN2003/001
2
Web File Transmission by Object Packaging
The following four subjects will be discussed
1. A new method for web file transmissions,
Object Packaging, is proposed
2. Performance comparison with existing HTTP
protocols
3. Advantages in using Object Packaging
4. Future work
LCN2003/002
3
Web File Transmission by Object Packaging

• The response time becomes the most significant
  concern

- Average users leave a web site after 8.5 second
delay

• Improving response time is not a simple task as
  it sounds

- Careful analysis of performance bottleneck
required

• If network protocol needs to be modified, it may
  not be
• soon to happen

- If hardware approach is used, how much will it
cost?
LCN2003/003
4
Web File Transmission by Object Packaging
A new file transmission method, Object Packaging,
is proposed for
? Efficiently reduce response time at a busy web
server
? Improve scalability of web servers
? Improve the throughput at intermediate routers
in a network
? No extra hardware resources required
? Network protocol transparent
LCN2003/004
5
Web File Transmission by Object Packaging

• Implemented directly on top of TCP

• TCP was designed for reliable bulk transmission

(but not for short-lived multiple sessions)

• For each file requested, RTT delay is
• required to set up a TCP connection

• As transmission rate increases,
• RTT becomes relatively longer

• For small files, RTT is relatively
• long

LCN2003/005
6
Web File Transmission by Object Packaging

• A protocol that is not solely dependent on TCP

LCN2003/006
7
Web File Transmission by Object Packaging

• In HTTP 1.1 Persistent Connection, boundaries of
  multiple web files are

- Still recognized by OS and file system
? FAT look-up is required for each file
- Still recognized by network/transport protocol
? Number of packets increase
? Protocol overhead increases (at end-hosts and
routers)
LCN2003/007
8
Web File Transmission by Object Packaging
Object Packaging
Transmit multiple files as single file
LCN2003/008
9
Web File Transmission by Object Packaging
HDD
LCN2003/009
10
Web File Transmission by Object Packaging
LCN2003/010
11
Web File Transmission by Object Packaging
LCN2003/011
12
Web File Transmission by Object Packaging
Client (a web browser) creates Object Package
Management Table (OPMT)
OPMT will be created based on the information in
Object Information Field
LCN2003/012
13
Web File Transmission by Object Packaging

• Compare the response time and CPU workload

• Evaluate the performance for HTTP 1.0, HTTP 1.1
  P.C. and Object Packaging

• Experiments using a LAN testbed

(Transmission of last packet) - (Arrival of
first request)
Server CPU Utilization (_at_100ms interval) for
above
(the last read function) - (the first read
function)
Client CPU Utilization (_at_100ms interval) from
the beginning to the end
LCN2003/013
14
Web File Transmission by Object Packaging

• 6-feet Cat-5 UTP Ethernet cable for both server
  and client hosts

• The LAN testbed was isolated from the Internet

• 100Mbps Fast-Ethernet

• The server and client hosts are both desk-top
  PCs (Pentium III 1.0GHz)

• Windows XP-Pro for both the server and client

LCN2003/014
15
Web File Transmission by Object Packaging

• HTTP 1.1 P.C. Server/Client Base Server/Client
  extended for P.C.

• Object Packaging Client Base Server with OBJP
  receiving function

LCN2003/015
16
Web File Transmission by Object Packaging
File Group
of sets in group
File Types
of files in a set
File Size
10
100
Text
4 KB
LCN2003/016
17
Web File Transmission by Object Packaging
(unit ms)
LCN2003/017
18
Web File Transmission by Object Packaging
LCN2003/018
19
Web File Transmission by Object Packaging
LCN2003/019
20
Web File Transmission by Object Packaging
LCN2003/020
21
Web File Transmission by Object Packaging
LCN2003/021
22
Web File Transmission by Object Packaging
(unit ms)
LCN2003/022
23
Web File Transmission by Object Packaging
LCN2003/023
24
Web File Transmission by Object Packaging
LCN2003/024
25
Web File Transmission by Object Packaging
LCN2003/025
26
Web File Transmission by Object Packaging
LCN2003/026
27
Web File Transmission by Object Packaging
0
20
0.48
LCN2003/027
28
Web File Transmission by Object Packaging
- OBJP reduced 92-56 of HTTP 1.0 (for 4-32 KB
files)
- OBJP reduced 64-15 of HTTP 1.1 P.C. (for 4-32
KB files)
- OBJP reduced 91-64 of HTTP 1.0 (for 4-32 KB
files)
- OBJP reduced 82-48 of HTTP 1.1 P.C. (for 4-32
KB files)
LCN2003/028
29
Web File Transmission by Object Packaging
- OBJP reduced 91-72 of HTTP 1.0 (for 4-32 KB
files)
- OBJP reduced 48-44 of HTTP 1.1 P.C. (for 4-32
KB files)
- OBJP reduced 97-93 of HTTP 1.0 (for 4-32 KB
files)
- OBJP reduced 57-64 of HTTP 1.1 P.C. (for 4-32
KB files)
LCN2003/029
30
Web File Transmission by Object Packaging
Major existing problems
? How can OBJP work with client-side caching?
? How can dynamic web pages (such as
CGI-generated) handled ?
? How will OBJP perform if multiple clients make
requests ?
? Repeat the experiments to improve the accuracy
of the output
LCN2003/030
31
Web File Transmission by Object Packaging
? We confirmed the significance of disk access
overhead at web servers
(especially for small files (4KB), the server
response time is one-tenth of HTTP 1.0)
? Object Packaging is efficient in reducing
server side response time
? Reduce server-side CPU work load
? Reduced client side overhead
? Does not require extra hardware resources
? Protocol/OS transparent
(but not years later)
LCN2003/031
32
Web File Transmission by Object Packaging
The End
LCN2003/032