Open GoogleCodeExporter opened 9 years ago
The current per-goroutine overhead with default setting of history_size is 680K.
With GORACE=history_size=0 it goes down to 455K.
I've landed a set of changes that bring it down to 222K. That's pretty much the
limit. 128K is occupied by thread trace (16K 8 byte entries), 64K is occupied
by thread vector clock (8K 8 byte entries). Plus some limited overhead for
everything else (including Go runtime as well).
I need to integrate fresh tsan runtime into Go tree so that it actually takes
effect.
Original comment by dvyu...@google.com
on 13 Feb 2015 at 5:46
FTR, I used the following program for tests:
http://play.golang.org/p/raXIeRPxQv
Original comment by dvyu...@google.com
on 13 Feb 2015 at 5:47
Original issue reported on code.google.com by
anacrolix@gmail.com
on 8 Feb 2015 at 1:21