To clarify the previous Tweet: MEM_LARGE_PAGES makes a linear 1gb write from cold start run at basically single-core memory bandwidth of the machine. Without MEM_LARGE_PAGES, it takes 2x the time.
-
-
Replying to @cmuratori
If I understand NT large page support correctly, large pages aren't swappable. So after VirualAlloc there isn't any more bookkeeping to do. There shouldn't be page faults on allocated large pages. This explains why you see peak bandwidth with large pages and no delay?
1 reply 0 retweets 0 likes
Replying to @badtakebrian
That is my assumption as well, at least. But the results are very telling: MEM_LARGE_PAGES takes 6ms in VirtualAlloc and then appears to have no overhead whatsoever. Without MEM_LARGE_PAGES, it takes 0ms in VirtualAlloc but ~60ms of page fault handling during first touches.
1:08 PM - 11 May 2021
0 replies
0 retweets
1 like
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.