[Openais] General Query
sdake at redhat.com
Mon Apr 24 12:10:38 PDT 2006
On Mon, 2006-04-24 at 14:29 +0100, Deepika Rani wrote:
> How to determine the service availability of a
> I mean suppose If a software is 99% available, it
> means its down time is around 36 hrs / yr. In order to
> prove that the software is really 99% available shud
> it be tested for one yr and then declare (or) are
> there any other approaches to determine the service
> availability of a software
To prove availability - hmm that is a tough question. Usually the best
that can be done is to model reliability of hardware in the system.
Then these reliability values will give you the MTBF. Unfortunately
software usually doesn't fail because of things like heat or chip
stress. Software usually fails because a) the user makes an error b)
the software has a defect. I suppose it is possible to model these MTBF
possibilities. Then these total system-wide MTBF can be taken to
produce the value of A depending on MTTR.
This provides a model under which availability can be predicted.
Proved? I've heard alot of schemes like "OK if I run this system for 3
months and failover 5 times in under 1 second, I've proved 5 9's
availability. I suppose this could provide some marketing support, but
in reality, it is very difficult to measure MTBF of software because
software usually doesn't follow the same failure model as hardware.
All of these schemes expect the failover operation to go smoothly. The
real challenge is to model what happens when the system does not
failover smoothly (because of a software defect in the application or
There are alot of good books on availability modeling. A quick amazon
search should find something helpful for you.
> HA Reader
> Yahoo! India Matrimony: Find your partner now. Go to http://yahoo.shaadi.com
> Openais mailing list
> Openais at lists.osdl.org
More information about the Openais