PT-server for LSUParc incredible slow

[fog]

The generation of the PT-server for LSUParc is extremely slow mainly due to the initial and final counting of the bases.

[meierh]

The performance of the initial procedure depends on the length of the alignment. As the alignment is more than three times longer in LSUParc than e.g. in SSURef, it takes more than three times as long.

[fog]

This ticket is more than 2 years old and completely outdated. Close it…

[westram]

• task still exist - only elmar (and so fog) has some hackarounds

[epruesse]

Yes and no.

At the time this ticket was created, the code that dumped the sequences from the db and removed the gap characters was very slow. I may have been the first to tinker with this, but the code has been optimized many times since then. r5918 is the latest change and is a big improvement especially with respect to code readability.

The runtime is still proportional to the alignment width, but — deep ARBDB magic aside — that's already optimal. Absolute runtime is probably also very close to optimal.

The hackaround Ralf mentioned caches the results of probe_compress_sequence on disk. It is only computed before the "arb_pt_server -build" and read directly from disk during "arb_pt_server -load". This is valid because changes in the always lead to a new build.

[westram]

by changes done in Nov 2013, see [11060]