Why so little parallelism?
|
From: BDH on 15 Dec 2006 21:42 > Think Dirichlet tesselation for a simple example of a foul task for > splitting; unless you know a fair amount about the data distribution, > you will always have problems. If you can't partition finding distance triplets, then you're storing your data wrong for exploitation of locality.
From: Nick Maclaren on 16 Dec 2006 05:30 In article <1166236948.765013.61290(a)t46g2000cwa.googlegroups.com>, "BDH" <bhauth(a)gmail.com> writes: |> |> > Think Dirichlet tesselation for a simple example of a foul task for |> > splitting; unless you know a fair amount about the data distribution, |> > you will always have problems. |> |> If you can't partition finding distance triplets, then you're storing |> your data wrong for exploitation of locality. And you are claiming that you can always work out an optimal data partitioning in 20 minutes? Regards, Nick Maclaren.
From: BDH on 16 Dec 2006 06:26 > |> > Think Dirichlet tesselation for a simple example of a foul task for > |> > splitting; unless you know a fair amount about the data distribution, > |> > you will always have problems. > |> > |> If you can't partition finding distance triplets, then you're storing > |> your data wrong for exploitation of locality. > > And you are claiming that you can always work out an optimal data > partitioning in 20 minutes? I claimed I could see *how* a *decent* partition *could* be done for the general case. Maybe it's arrogance, but I consider myself a damn fine algorithm designer, and some days am miffed at how little good it's done me.
From: Nick Maclaren on 16 Dec 2006 06:40 In article <1166268405.965668.182760(a)n67g2000cwd.googlegroups.com>, "BDH" <bhauth(a)gmail.com> writes: |> |> > |> > Think Dirichlet tesselation for a simple example of a foul task for |> > |> > splitting; unless you know a fair amount about the data distribution, |> > |> > you will always have problems. |> > |> |> > |> If you can't partition finding distance triplets, then you're storing |> > |> your data wrong for exploitation of locality. |> > |> > And you are claiming that you can always work out an optimal data |> > partitioning in 20 minutes? |> |> I claimed I could see *how* a *decent* partition *could* be done for |> the general case. OK. Now please put up or shut up. Let's see a write-up of your design. In other words, publish OR be damned as a blowhard. |> Maybe it's arrogance, but I consider myself a damn fine algorithm |> designer, and some days am miffed at how little good it's done me. Well, some of us are considered quite good algorithm designers by other people, and it hasn't done us a lot of good, so why should you expect your self-appreciation to matter? Regards, Nick Maclaren.
From: BDH on 16 Dec 2006 07:09
> |> I claimed I could see *how* a *decent* partition *could* be done for > |> the general case. > > OK. Now please put up or shut up. Let's see a write-up of your design. > In other words, publish OR be damned as a blowhard. That's an unsolved problem? > Well, some of us are considered quite good algorithm designers by other > people, and it hasn't done us a lot of good, so why should you expect > your self-appreciation to matter? Don't, just the reason for some of my posting. |