[squeak-dev] OmniBase/Magma - how big??
goran at krampe.se
goran at krampe.se
Fri Feb 29 07:49:09 UTC 2008
Hi!
(ccing Magma list)
"Bill Schwab" <BSchwab at anest.ufl.edu> wrote:
> Hello all,
>
> Any success or horror stories with OmniBase and/or Magma? Can you
> report that either worked well at or got into trouble beyond any
> particular size (either in object count or bytes of data)? Other
> comments?
>
> Bill
>From the Gjallar project we can report that Magma works really nice. We
have had perhaps 2-3 buglets and Chris fixed them all quickly and
promptly. IIRC they were related mainly to the query stuff which Chris
pushed out fast because of our popular demand - so he was definitely not
to blame. :)
Generally I would say that the only "issue" we had was slow(ish) read
performance. So if you write an "interactive" application with Magma you
need to take two things into account:
1. Since Magma is an ODB it builds a cache of objects into your session.
So when/if the session is cold you get punished with load times. But
when it is hot they are zero. Essentially what this means in Gjallar is
that when a user logs in things will take a bit of time the first time
she does something. But from then on things go much faster and since a
Gjallar user typically is logged in for a long time - typically all day
- it works fine.
2. Even with a warm session (with lots of cached objects) you can still
get punished with longish load times. This can be remedied a LOT with
setting up good read strategies.
IMHO I think Magma would SHINE if we somehow could boost the
serialization/deserialization mechanism in it. Because I suspect this is
where the bottleneck is regarding the above. An Exuperified Magma could
be the answer - or some other HARD approach - I dunno. Chris?
regards, Göran
More information about the Squeak-dev
mailing list
|