[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: scaling object databases

> > Some estimates from the files and prototyped code:
> >
> > a) 1 million users
> > b) each user performs 20 transactions/day
> this means you will have around 10,000 transaction per second (!) in the
> peaks...

What's your math here? 1,000,000 users @ 20 transactions/day is 20,000,000
transactions /day

# of seconds in a day = 60 * 60 * 24 = 86400

20,000,000 / 86400 ~= 231 transactions / second

Peak traffic of 50 * average traffic seems unlikely. Based on my experience
with the web, peaks of 5 * average are only associated with things like a
report that comes out at the same time each day, etc, etc.

> > c) half of the transactions are read-only
> > d) total data amounts to 50 million objects
> > e) average transactions affect 10 objects each, either reading
> or writing.
> > f) physically, the overall system is very centralized. If there are
> > clustered servers, they will all be in racks in the same
> hosting location.
> > g) uptime is a priority, so the system must have failover capability.
> ... IMO ozone is not suited for that app. I wonder what kind of
> Java based (or
> other) system is able to handle these requirements anyhow.

Having refigured the transaction rate above, still unsuited for Java based
on overall requirements? What sort of transaction rate would you expect
Ozone to be up to? (Depends on machine speed, et al, of course, but

Published benchmarks for Ozone would be really really useful; have a look at
http://www.tneumann.de/en/SOD.html, an ODB that provides some benchmarking
for guidelines.