My work group is just starting the processing of finding a ODBMS product
to replace our in-house developed persistence layer for our high volume
data processing.
I greatly value the experiences & feedback that other users/customers
may be willing to share. Any feedback would be greatly appreciated.

We have the following requirements:
-C++ API
-handle large volumn of data: 500,000+ objects of different types/sizes
-ability to checkpoint programmatically (API provides means to
initiate/verify checkpointing)
-embedded (objects can be updated multiple times via memory aliasing ie.
database gives the app a reference that can be manipulated after the
initial insert)
- reentrancy & level 1 MT-safety
- ability to add new object types and attributes to existing objects on
the fly (the additional attributes may be added by adding existing db
type elements to an exisiting variable length attribute array).
- the 'standard' ACID characteristics...
- flushing to disk can be controlled programmatically (we would often
[but not always] prefer the speed over the risk of data loss between
flushes)
- selection of objects via the identity values (ie. primary key of each
type).

Nice to have features:
- java API
- API choice of locking/concurrency strategies (population phase is done
by a single process but subsequence process phases may require
multiple/concurrent access)
- ability to query objects via non-identity attributes (indexes or table
scan like lookups).
- ability to define & generate non-identity indices (especially after
the initial database population phase).
- import/export utility

Some of the products we identified late last week are:
eXtremeDB from MCObject
Objectivity
Fastobjects
Intersystem's Cache
Alpha Five Version 5
Relexus' relX embedded
Equi4
Encirq

Any feedback (good & bad) on any of these products or others would be
greatly appreciated.

OO/RDBMS mapping products may be useful but only if they are super fast.
We'd like to first try to avoid the mapping avenue unless it is the only
choice.

-steve
shespelt@_remove_me (AT) bear (DOT) com

Steve,

You may also want to take a look at Versant and Object Store from
Object Design (now Progress Software). Both of these systems will
be able to meet the requirements you have outlined below.

--Corey

"Steve Hespelt" <fooz (AT) tellurian (DOT) net> wrote

"Corey Brown" <corey (AT) spectrumsoftware (DOT) net> schrieb im Newsbeitrag
news:WjtTb.10251$7I3.9230 (AT) bignews4 (DOT) bellsouth.net...
I do know ObjectStore a bit (although my knowledge is a bit rusty) and as
far as I remember schema migration was not as easy as "on the fly". That
might have changed.

Other than that ObjectStore gives you quite a lot of control on how
objects are aligned in on memory pages and thus you can tune performance
and concurrency quite well. There's even a mode where multiple clients
can concurrently read the old contents of pages that are about to be
updated.

What about Objectivitiy? AFAIK it's a page server, too, thus it should be
quite fast and I've heard of at least one Terabyte database implemented
with Objectivity.

Regards

robert

Steve Hespelt <fooz (AT) tellurian (DOT) net> wrote

Don't forget Matisse (http://www.matisse.com).