search

MDAnalysis / scipy_proceedings

SciPy conference proceedings: MDAnalysis paper
Other
0 stars 4 forks source link

paragraph on trajectory offsets for random access #31

Closed orbeckst closed 8 years ago

orbeckst commented 8 years ago

Could I please get a paragraph outlining our approach to speeding up trajectory access?

This is pretty unique in the way that we do it. @mnmelo ?

richardjgowers commented 8 years ago

Also worth mentioning we do this for ascii files too which allows us to seek them too

On Mon, 30 May 2016, 20:18 Oliver Beckstein, notifications@github.com wrote:

Could I please get a paragraph outlining our approach to speeding up trajectory access?

This is pretty unique in the way that we do it. @mnmelo https://github.com/mnmelo ?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/MDAnalysis/scipy_proceedings/issues/31, or mute the thread https://github.com/notifications/unsubscribe/AI0jB3G8z89SQScqogNWWj3Vsdcqyy75ks5qGzgagaJpZM4Ip_Yv .

kain88-de commented 8 years ago

That should go into basic usage right.

@orbeckst I'd appreciate if we do this days before the deadline next time and not hours!

kain88-de commented 8 years ago

the basic usage already hints at it but doesn't mentiong that it's super awesome that we actually do that.

orbeckst commented 8 years ago

On 30 May, 2016, at 12:22, kain88-de notifications@github.com wrote:

@orbeckst I’d appreciate if we do this days before the deadline next time and not hours!

I hear you.

If there are things that you can’t do today: that’s fine, just say in the issue. It’s not a problem. I know that this is all late. Pretty much everything is a “nice to have”. I can always put a stub in the draft. It’s generally just better if someone wrote something who actually did the work.

The conference paper is really not such a big deal. If it’s not perfect then that won’t be the end of the world. As long as it looks fairly reasonable (and it already does) we’ll have an opportunity to fix-up until July 5.

kain88-de commented 8 years ago

I would like to add this when we have a PR to unify the offset store and load behavior. Currently we only support it for XDR files (the PDB file generates them at every load). After that we can go into detail how this is solved in a general way in MDAnalysis

orbeckst commented 8 years ago

Good point @kain88-de . Then I'd just want to say that stored offsets (which originated in LOOS) are a viable solution to providing random access for any trajectory format. We have already implemented it for the Gromacs formats and will soon be adding this to all formats.

The full MDA paper will then have a more in-depth discussion of stored offsets and benchmarks.

kain88-de commented 8 years ago

Yes you can say that. I'd appreciate it if you could add that because I'm not sure where it would fit best.

orbeckst commented 8 years ago

Ok, will do. Just add comments here (goes for @mnmelo , too).