Persistence & DTSTTCPW: ZODB clone?

Scott A Crosby crosby at qwes.math.cmu.edu
Wed Feb 6 12:28:01 UTC 2002


On 6 Feb 2002, Cees de Groot wrote:

> Jimmie Houchin <jhouchin at texoma.net> said:
> >Would this be a reasonably faithful port or would you attempt to make
> >changes?
> >
> It wouldn't be a port at all, just something with the same design.
>
> >It would be interesting to benchmark the two and see how they compare.
> >
> ZODB is likely to fare better, because the serialization is done in C code and
> that's likely to be the bottleneck.
>

It could be done nominally as fast as an ImageSegment, or faster.

Put a dirty bit into the object header. Make a variant of
ImageSegment's that act as normal, but only dump objects if they're
so-marked as dirty.

It could be made faster, have a variant of GC that uses extra header bits
(so-as to not collide with normal GC), where the user defines the roots
that are used and what objects to scan and/or ignore.

A variation of this requires that if any object is 'dirty', then some path
from the root to that object must also be marked dirty. This would be
fastest, scan all dirty objects from the roots downward, but would require
manual support to build such paths.

All of these also need some unique identifier for each object.

> First I need to attach Scott's FTS engine to the Squeak mailing list
> archive

Its available, but its not been prepared or tested on the module system,
so I'm holding back on the release. (And with that pig to swallow, it
won't get through anytime soon anyways.)

Scott




More information about the Squeak-dev mailing list