Transcript PowerPoint

Reducing Network Latency
Paul Johnson
CPSC 550
21 APRIL 05
Using an Intelligent Service to
Determine the Cheapest
Communications Path
Problem



The von Neumann bottleneck is made
worse by networking
The networked information is just a new
level in the memory hierarchy
There is evidence that the latency for
communication is not consistent, even
when looking for the same server
Objective

Create an intelligent service capable
of reducing the overall
communication latency by providing
the best known routing path to the
operating system on demand.
Connectionless Communication

Observing out of order datagram
reception over a local network
Monitoring Data Transfers (1)

PING returns the latency when
communicating with another computer
Monitoring Data Transfers (2)

PING can help recognize long
response times
Monitoring Data Transfers (3)

TRACERT displays the hops from
point to point on the internet
Monitoring Data Transfers (4)

TRACERT again goes to a new destination,
and takes different hops
The Windows Way

ROUTE PRINT shows the METRIC used to
determine which route to use sending data
Proposed Solution (1)



Addition of a service to monitor possible
routes to common servers
The service would provide information to
the OS on demand regarding the best
path to a destination
Nodes at hop destinations could re-route
the packet based on more complete
information
Proposed Solution (2)

The service would know the best route
from User to Server
Possible Implementation

Runs in the background, taking advantage of unused clock
cycles

AI Expert System using rules based on consistency, time of
return, and number of hops required

Testing packets created by the service can be dropped at
any given time if “real work” becomes available

Partial path to the destination is maintained within the
message

Entirely theoretical design, may require new types of
hardware to support the service
Final Remarks



Based on time of day, server load, and
priority, each node may begin to have
favorite paths – reducing contention for
resources
This is a dynamic solution, capable of
handling changes to networks
The solution handles hardware failures
without additional intervention
References


Coulouris, George; Dollimore, Jean; and Kindberg, Tim. Distributed
Systems: Concepts and Design, 3rd Edition. New York: Addison-Wesley,
2001.
Hennessy, John L. and Patterson, David A. Computer Architecture: A
Quantitative Approach, 3rd Edition. San Francisco, California: Morgan
Kaufmann Publishers, 2003.

Giarratano, Joseph. Expert Systems: Principles and Programming, 3rd
Edition. Boston, Maryland: PWS Publishing Company, 1998

Jackson, Peter. Introduction to Expert Systems, 3rd Edition. New York:
Addison-Wesley, 1999.

Keogh, Jim. The Essential Guide to Networking. Upper Saddle River, New
Jersey: Prentice-Hall, Inc., 2001.

Miller, Mark. Internet Technologies Handbook: Optimizing the IP Network.
Hoboken, New Jersey: John Wiley & Sons, 2004.