[MOBY-l] Re: moby-l digest, Vol 1 #71 - 1 msg

[Jason E. Stewart]

mwilkinson at gene.pbi.nrc.ca writes:

> > As a service provider, this also dealt with other troubling aspects of
> > allowing clients to send lists of objects: (i) Do you have to consider
> > whether a client sends a list that is too long? e.g. all ID's in GenBank?
> > (ii) Does constructing the response object for a large list of input
> > objects consume prohibitive amounts of cpu/disk/memory? e.g. BLAST
> > reports.
> 
> Brian Gilman would say "spit out the answers a bit at a time"...

Resource allocation is a policy best left up to the providers. If they
get requests that are too large (and they will), they should have a
queuing mechanism.

> >From my perspective, sending all ID's in genbank is a perfectly valid
> (though horrific) query, and we shouldn't prevent people from doing this.
> The disadvantages of sending one request at a time are numerous, not the
> least of which is the overhead of SOAP itself, which is painfully slow...
> I'd rather claw my eyes out than have to do 1000 identical SOAP queries one
> at a time!  Another problem is that some services will *require* a list...
> e.g. sequence alignment services.  I don't want to have to define special
> "list" objects on top of it all :-P

Urm. Alignment is a good example.

Mayhaps the MIB block should be ressurected from an early grave to
serve the role of identifying which bits of the output are associated
with what bits of the input?????

jas.