On Thu, Aug 20, 2015 at 11:47 AM, Tobias Pape <Das.Linux at gmx.de> wrote:

> Hi Mariano,
>
> On 20.08.2015, at 16:41, Mariano Martinez Peck <marianopeck at gmail.com>
> wrote:
> > […]
> > Is this similar to Tanker in the way that classes and methods are binary
> serialized with Fuel? (no compiling)
> > Or you just used Fuel instead of the current usage of MCDataStream? If
> this is the case, then you would still be compiling at load time etc. We
> already did this experiment long ago and we didn't find a significant
> speedup.
> >
> > What were your findings?
> […]
> Well that was 4 years ago:
> http://forum.world.st/Monticello-enhancing-td4179277.html
> And I don't remember too well, just my feeling that it was a lot faster by
> replacing
> the DataStream blob with a fuel blob (so, no, not like Tanker).
> (No actual numbers written down, tho, sorry)
>
>
OK, that's what I imagined. Yes, we did the same experiment too. As said,
we haven't seen an speed up there because even if Fuel would be 100x faster
writing that blob, the writing of that blob was insignificant in a
Monticello load. Most of the time when loading with Monticello goes into
the parsing/compiling, the class building thingies (may involve became: to
update instances), the notifications, etc..

In fact, one of the things we found is that:
-  just having a bulk notification (and update all the observers) would
speed up things (have you ever tried loading with Monticello with
TestRunner opened???? takes a lifetime!)
- Spur would help us with the lazy become when updating instances

So probably maybe with those 2 things we already see an speedup even if we
still compile from source...



-- 
Mariano
http://marianopeck.wordpress.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20150820/45bba066/attachment.htm