Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: performance question: sum values vs. store calculations

Re: performance question: sum values vs. store calculations

From: Alexander I. Doroshko <aid_at_grant.kharkov.ua>
Date: 1 Sep 1999 08:29:34 GMT
Message-ID: <01bef453$b00f2cc0$190114c1@sister.grant.UUCP>


We had such an experience in a banking system. It resulted in negation of the approach mentioned, and now we store the current accounts residues as well as atomic transactions.
The performance issues weren't the main ones, but that of minimizing of locking, consistency preserving and code simplicity.
As to the performance issue, it became very important while, for example, the work for a great period of time. So it was necessary to store the day's residues (but we have several millions of transactions per year over some accounts).
But then there must be a great difference while working with current and archive data, which is very unpleasant.   And the main issue was the convenience of consistency preserving - you lock a row with the account and make a transaction, and nobody can make his own transaction at the same time. It isn't so simple while storing only atomic transactions.
--
 Alexander I.Doroshko, aid_at_grant.kharkov.ua

Todder <toddbrasseur_at_my-deja.com> wrote in article <7qgq1s$n0m$1_at_nnrp1.deja.com>...
: Sorry, not much help here.
:
: We have the similiar question. We currently have an application
: (written in RPG on AS/400 and in CLIPPER (dbf) which both take the
: approach of not saving any data which can be calculated. This has
been
: very successful and we are very happy with performance. Our largest
: table is only around 100,000 records but the calculation process is
: very involved.
:
: We are now moving to a relational database. We are approx. 90%
: complete with the development. Although the response time isn't
quite
: as fast as it was (with DBF's) it is still acceptable. We also
believe
: that computers will continue to be faster and faster.
:
: Todd @ COMPASS
:
:
:
: In article <7qfkq5$si9$1_at_nnrp1.deja.com>,
: erik_ostermueller_at_my-deja.com wrote:
: > The legacy banking system that we are rewriting stores the results
of
: many
: > calculations in tables. For instance, when a balance is calculated
: for an
: > account for a new transaction, the new balance is stored on the
: record for
: > the account. For flexibility and maintainability reasons, we would
: prefer
: > to store only the atomic transaction data, and derive/calculate the
: balance
: > (and other calculations) when requested.We know that the tradeoff
for
: this
: > decision is a performance hit, but don't know how big it will
be.Each
: > calculation will involve about 50-100 records from a total of 1.5
: billion
: > records. This will be a 24x7 system that expects sub-second
response
: time.
: > Does anyone out there have some experience with this?
Received on Wed Sep 01 1999 - 03:29:34 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US