These are chat archives for atomix/atomix

16th
Jan 2016
Richard Pijnenburg
@electrical
Jan 16 2016 10:28
Hahaha just a bit. lol
Richard Pijnenburg
@electrical
Jan 16 2016 22:17
Should be more sync now lol.
Jordan Halterman
@kuujo
Jan 16 2016 22:19
:-)
Richard Pijnenburg
@electrical
Jan 16 2016 22:20
How's your weekend so far ?
Jordan Halterman
@kuujo
Jan 16 2016 22:21
good good
Just going through more little bugs and documentation this weekend.
Richard Pijnenburg
@electrical
Jan 16 2016 22:22
Nice
Jordan Halterman
@kuujo
Jan 16 2016 22:22
the website docs are way out of sync
Richard Pijnenburg
@electrical
Jan 16 2016 22:22
Hehe yeah
Jordan Halterman
@kuujo
Jan 16 2016 22:22
updating them over the next week so we can release a release candidate
doing some performance improvements along the way
Richard Pijnenburg
@electrical
Jan 16 2016 22:25
Very nice. Looking forward to mess around it further.
Jordan Halterman
@kuujo
Jan 16 2016 22:39
It’s only going to progressively get better no doubt. Just bug fixes and performance improvements. Once the website is updated and a release candidate is pushed we’ll try to actually get more users. Between now and then there are a lot of little improvements to be made. For instance, doing some performance improvements on the log today. Right now is stores some data fairly inefficiently - namely log terms. Basically, the same term is usually repeated over and over for each entry, and terms always increase and never decrease in the log, so that makes it a good candidate for some improvements. Basically, the log should just log a term once and all subsequent entries should inheret that term. There’s also some opportunity to do less deserialization on servers. Right now everything is unnecessarily deserialized every time the log is accessed. For example, an entire entry is deserialized just to read the term. That’s unnecessary. There’s also some caching that can be done in the log to prevent writing an entry to disk and then reading it to send it to followers. It should also be modified to map only recent segments into memory when using StorageLevel.MAPPED. I think the log is the obvious performance bottleneck and that’s where huge performance improvements can be made fairly easily.
Richard Pijnenburg
@electrical
Jan 16 2016 22:42
Yeah. Especially with the task queue it's easy to see there are improvements that can be done.
Jordan Halterman
@kuujo
Jan 16 2016 22:42
totally
there should be a lot of pretty easy ones that will result in significant improvements without having to do a bunch of benchmarking and crap
Richard Pijnenburg
@electrical
Jan 16 2016 22:44
Hehe yeah.