[Seaside] Goods

[Chris Muller]

I would just like to clarify, the Magma set up is "one-click" and
always has been.  Ramiro was describing his experience loading the
"queries beta" version a couple of weeks ago, which was not one-click. 
That version was finished and announced last week, it is no longer in
beta, is therefore now a one-click install.

I would also like to clarify the performance comment.  Ramiro, I
profiled your 

  NYSE instance getValuesFor: 'aapl' from: ('1/1/2004' asDate) to:
('1/1/2006' asDate)

which addeds 504 objects and would like to clarify a couple of points
about the 33 second profile:

  - 5.4 seconds was spent getting the data from Yahoo
  - 18.1 seconds spent doing 504 queries on the MagmaSet to determine
if the object being added already exists.  (NOTE 6 of which in
DateAndTime>>#asUTC).
  - 10 seconds for the actual commit

MagmaSets are expensive to add to, because every object requires it to
query to see if that object already exists.  You could use a
MagmaCollection here and would see the time cut in half, 5.4 seconds
for Yahoo and 10 seconds for the commmit.

This is why I cringe at public declarations about performance.  One
persons 260 objects is different than anothers.  I think the only
meaningful way to describe or compare performance is with 1)
well-explained execution contexts (hardware, etc.), 2) detailed
descriptions of the task being measured, and 3) what performance-tuning
considerations (which, in Magma's case, can have a 10X improvement)
were done.

MagmaBenchmarker attempts to address these items:

  1) the execution context includes computer, cpu, speed, memory, and
disk
  2) the benchmarking methods names are descriptive in what they do
  3) the implementation of the methods can be examined to see what
performance tuning mechanisms were used, if any

To see this output for the latest Magma:

  http://minnow.cc.gatech.edu/squeak/5606

I certainly don't want to proclaim Magma is blazing fast, but the
MagmaBenchmarker on the current version just demonstrated, on my circa
2004 laptop, insertion of 1000 objects (100 at a time, 10 commits) into
a MagmaCollection in 4 seconds flat (5 without WriteBarrier).

Understanding performance requires context, lots of context..

 - Chris



> Magma
> ======
> 
> Good:
> 	New cool query syntax (no one else has this).
> 	Excellent support from Chris Müller and list.
> 	Clean syntax
> 	Indexes for various classes (Dates, Strings, Numbers, etc.)
> 	ReadStrategies
> 	Native Smalltalk
> 	Integration package to use with Seaside.
> Bad:
> 	Hard to set up (the latest beta version with queries at least)
> 	Insert & update performance (inserting  ~260 items on a magma  
> collection takes >30'' on a single commit)
> 	
> 
> 
> GOODS
> ======
> 
> Good:
> 	Very simple to set up
> 	Cleanest / less intrusive syntax of all ODBs I've seen
> 	Pretty fast on inserts (on my not so large data sets at least)
> 	Distributable / multi-language
> Bad:
> 	Maintained by a single guy (I'm always scared of this... what if he 
> 
> decides to run a restaurant?)
> 	Not native Smalltalk
> 	No indexes (must use Avi's BTree)
> 	No queries
> 
> 
> OmniBase
> ========
> Good:
> 	Seems to have been widely tested in several smalltalks
> Bad:
> 	Locking problem on Macs and Linux with Squeak has not been resolved.
>  
> Hence, if you use Squeak, Windows is the only alternative.  I've  
> tested it yesterday on my Mac and the locking issues were still  
> there.  I don't know if it is still this way under Linux.
> 	Ugliest syntax (for me, at least)
> 	No query language.
> 	Licencing? (don't really know about this)
> 
> 
> But it would be great, at least for me, to hear what others been  
> through when they decided to implement an ODB.
> Thanks
> 
> 
> r.