Browse Prior Art Database

Dynamic Adjustment of Timer to Terminate Multicast Queries

IP.com Disclosure Number: IPCOM000108816D
Original Publication Date: 1992-Jun-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 2 page(s) / 92K

Publishing Venue

IBM

Related People

Gopal, PM: AUTHOR [+2]

Abstract

Disclosed is a method of setting a timer associated with client queries in client-server distributed computing, based on the measured response time for prior successful queries. The method selects an appropriate initial timer value, then dynamically tunes it as network conditions change, so that good performance may be obtained.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Dynamic Adjustment of Timer to Terminate Multicast Queries

       Disclosed is a method of setting a timer associated with
client queries in client-server distributed computing, based on the
measured response time for prior successful queries.  The method
selects an appropriate initial timer value, then dynamically tunes it
as network conditions change, so that good performance may be
obtained.

      Client-server interactions typically begin with an optimistic
first phase where a reply is first sought from only the servers
possessing pertinent data.  If the client receives no replies, it
begins an exhaustive second phase, repeating the query with an
indication that all servers should reply, whether or not they have
the data.  The client typically uses a local timer to determine when
to end the first phase and begin the second.  Several reasons for not
receiving a reply before timer expiration are:
       .   None of the servers has the queried data; consequently, no
replies were sent.
       .   Replies were sent, but were lost by the communication
medium.
       .   Replies were sent, but did not reach the client before its
timer expired, because the value was too small.

      The client's goal is to avoid the second phase, since it costs
more (in terms of messages) than the first.  A long timer value
increases the probability of successfully terminating the optimistic
phase, but also increases response time by allowing for delayed
replies.  On the other hand, a short timer value decreases phase 1
response time at the expense of successful termination, and may
ultimately cost more if it causes more exhaustive queries.  The timer
value should be selected carefully to balance these two conflicting
goals.  Constantly changing network conditions make this problem
difficult, since the length of time for a message to traverse the
network varies as a function of queuing delays and available paths.

      The initial timer value is the m...