This change has significant performance implications. bufferv2 is reference
counted and pooled, which alleviates heap/GC pressure. Below are the results
from running the iperf benchmark.
HEAD:
BenchmarkIperf/operation.Upload-16 1552 ns/op 46.6GiB total allocations
BenchmarkIperf/operation.Download-16 1114 ns/op 68.6GiB total allocations
w/ change:
BenchmarkIperf/operation.Upload-16 1139 ns/op (-27%) 1.41GiB total allocations (-97%)
BenchmarkIperf/operation.Download-16 753.2 ns/op (-33%) 706MiB total allocations (-99%)
PiperOrigin-RevId: 462453185
When transmitting packets we only need to notify if the peer is not
already processing packets. sharedData region is used to enable/disable
notifications and the peer will disable notifications when its actively
processing packets and enable notifications just before it goes to
sleep waiting on packets. This allows more efficient transmit as the
sharedmem endpoint does not need to notify on eventFD and incur an
expensive host systemcall when the peer is already awake.
PiperOrigin-RevId: 406018843
The same create/write/read pattern is copied around several places. It's easier
to understand in a package with names and comments, and we can reuse the smart
blocking code in package rawfile.
PiperOrigin-RevId: 401647108
Nayana BidariandgVisor bot
Kevin KrakauerandgVisor bot
Lucas ManningandgVisor bot
Bhasker HariharanandgVisor bot