I was wondering whether it would be possible to to achieve a better performance by adding multithreaded code to some GS hot areas.
I can think of at least two:
- compressed backups. AFAIK Most of the time is spent in the gz data compression. It probably would make sense distributing compression computation among multiple cores (maybe with the number of the cores being parametrizable).
- mark for collection: we tuned for speed MfC the best we could on a machine with huge amounts of memory. The bottleneck in this situation seems to be the processing power needed (the gem in MfC needs 6hours at 100% usage.of a single core). Maybe GS could be improved with multithreaded MfC code if possible.
Please anyone feel free to add other ideas on the topic :)
It could be also Interesting to do the same for SystemRepository>>listInstances: and SystemRepository>>listReferences: