On 05/18/2013 03:27 PM, Brian Anderson wrote:
The new scheduler is still far from complete, but there is enough in
place that we can start to get a sense for its performance
characteristics. I've done a simulation of making message-passing
calls into a layout task and have some preliminary numbers.
Some additional notes that may be worth mentioning:
* Both the PortOne and ChanOne implementations contain an extra
allocation to work around a bug in Rust by-value self paramaters. That's
4 extra allocations per iteration in this test.
* The variance with the old scheduler was very high, and I saw times
from 10000 ns to 30000 ns. The new scheduler's time was always ~2200 ns.
_______________________________________________
Rust-dev mailing list
[email protected]
https://mail.mozilla.org/listinfo/rust-dev
