Ring buffers #12

New Issue

Jef Roosens · 2024-03-09T14:13:18+01:00

Jef Roosens commented

2024-03-09 14:13:18 +01:00

The read/write buffers for requests currently using memmove to move data to the front of the buffer, so the back can be reused for a subsequent read/write call. This works, but isn't the most efficient it can be, as it requires moving the data after every syscall.

An alternative to this would be using circular buffers. One disadvantage of this would be that syscalls could get split up into two (e.g. one read to fill the end of the array, then another to fill the part in front) but this performance decrease might not be too much of an issue, assuming the switch to circular buffers compensates for it. If we'd really like to go all out, we could instead start using recvmsg and sendmsg which can be instructed to write to multiple buffers. This way, we could write to the circular buffer using a single syscall.

The read/write buffers for requests currently using `memmove` to move data to the front of the buffer, so the back can be reused for a subsequent read/write call. This works, but isn't the most efficient it can be, as it requires moving the data after every syscall. An alternative to this would be using circular buffers. One disadvantage of this would be that syscalls could get split up into two (e.g. one read to fill the end of the array, then another to fill the part in front) but this performance decrease might not be too much of an issue, assuming the switch to circular buffers compensates for it. If we'd really like to go all out, we could instead start using `recvmsg` and `sendmsg` which can be instructed to write to multiple buffers. This way, we could write to the circular buffer using a single syscall.

Jef Roosens added the

enhancement

label 2024-03-09 14:13:18 +01:00

Sign in to join this conversation.