Postnuc speed #30
Postnuc speed #30
Conversation
Previously each `Node` in the alignment graph was wasting huge amounts of memory. Each `Score` structure had seven bytes of padding at the end. Thus, roughly a third of the memory occupied by a `Node` is unused. With the new format, values are tightly packed making a `Node` smaller. For an unrepresentative sample of two E. coli genomes, the new postnuc is 10% faster and uses 50% less memory. The former might be due to better caching.
|
I came across this via https://blog.kloetzl.info/mummer-the/ via Twitter. As an occasional user of mummer for genome assembly alignment as part of QA, I'd welcome the reported typical 1/3 less memory usage and 20% speed increase. |
|
Ouf this is quite old |
|
Is there any reason why this hasn’t been merged? Is this code base being maintained any more? |
|
@kloetzl Thanks for the patch. I am still using Mummer3 and I was wondering if there was an easy way to backport the postnuc changes? I can probably figure it out given enough time and your PR but any tips would be appreciated. |
|
I'm afraid I don't have my mummer3 patch any more. The changes to the structure should be the same, but the changes to the algorithm might be a bit more verbose. Mummer3 is using an older C++ standard so everything is slightly more tedious. Good luck! |
|
Thanks for replying! |
5 participants
Hi,
I am happy to see that there is still work done on MUMmer. Here is a patch, that I had lying around for a while. It greatly increases the performance of postnuc. This pull request also includes the patch to issue #29.
Best,
Fabian