Browse Prior Art Database

Distributed Load-Sharing Scheme Based on Non-linear Regression

IP.com Disclosure Number: IPCOM000105114D
Original Publication Date: 1993-Jun-01
Included in the Prior Art Database: 2005-Mar-19
Document File: 4 page(s) / 147K

Publishing Venue

IBM

Related People

Leff, A: AUTHOR [+2]

Abstract

A strategy is proposed that addresses the effect of information lags on load sharing using non-linear regression in a distributed database environment. It explicitly takes the information obsolescence into account when making routing decisions, and weights older information less heavily than more up-to-date information. Databases are partitioned across multiple sites and transaction can be executed at any site where for non-local database access, database call needs to be shipped to the remote site owning the database. This is referred to as remote database call. Routing an incoming transaction to different sites will incur different amounts of remote calls. This site affinity further complicates the problem.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 32% of the total text.

Distributed Load-Sharing Scheme Based on Non-linear Regression

      A strategy is proposed that addresses the effect of information
lags on load sharing using non-linear regression in a distributed
database environment.  It explicitly takes the information
obsolescence into account when making routing decisions, and weights
older information less heavily than more up-to-date information.
Databases are partitioned across multiple sites and transaction can
be executed at any site where for non-local database access, database
call needs to be shipped to the remote site owning the database.
This is referred to as remote database call.  Routing an incoming
transaction to different sites will incur different amounts of remote
calls.  This site affinity further complicates the problem.  Because
dynamic state information is not immediately available to the various
sites, decisions which are naively made based on the obsolete
information that is currently available will be non-optimal.  In this
environment sites must solve the problem of whether to select
alternative sites to process incoming transactions, given that the
information on which the decision is based exhibits varying degrees
of obsolescence.  We present a distributed load sharing strategy that
explicitly takes the obsolescence into account when making routing
decisions, and weights older information less heavily than more
up-to-date information.  The bias against the information lag will be
differentiated based upon the site affinity as contrast to the scheme
considered in [1].

      The proposed strategy consists of three steps.  In the first,
each site estimates the response time of incoming transactions.

Then, as transactions are processed and leave the system, sites track
the transaction's actual response time, and record its critical
characteristics with respect to the factors discussed before .  This
enables the system to monitor how well actual response time conforms
to the estimate.  Finally, the system adaptively adjusts future
estimates by periodically recalculating coefficients that are used to
weight the factors in response time estimation.

   Assume that there are  K  classes of transactions.  Let R sup ki
sub a  represent the actual response time of the i-th  class k
transaction,
 R sup ki sub e  its estimated response time similar to that under
the MRT strategy in [2]; X sup ki =1 if a remote routing decision is
taken for that transaction (0 otherwise); and let  T sup ki be the
difference between the current time and the time of the last
information update from the site to which the transaction is routed.
It seems reasonable that a transaction's response time, when routed
to a specific site, will depend on (a) the MRT estimate which
considers a transaction's database-site reference locality and
short-term load at the destination site, (b) whether the site is the
transaction's preferred site, which is the site resulting in the
smallest number o...