Actions
Bug #4499
closedVerify we set TCP_NODELAY socket option where needed
Start date:
04/17/2020
Due date:
% Done:
100%
Spec Reference:
Description
Recently it was spotted that we were not setting TCP_NODELAY option in OML and RSL connections, which means some messages being sent can be delay by considerable amounts of times. This kind of behavior is not beneficial in the context of signalling we do where latency is far more important than amount of data / network performance.
This ticket is a reminder to look for all protocols we implement that use TCP and verify whether it makes sense to enable TCP_NODELAY sockopt. Some I can think of include:- Abis (bts and bsc side)
- OML (bts and bsc side)
- GSUP (client and server side).
diff --git a/src/input/ipaccess.c b/src/input/ipaccess.c index e1936af..4d21416 100644 --- a/src/input/ipaccess.c +++ b/src/input/ipaccess.c @@ -271,6 +271,7 @@ static int ipaccess_rcvmsg(struct e1inp_line *line, struct msgb *msg, struct e1inp_ts *ts; struct osmo_fd *newbfd; struct e1inp_line *new_line; + int on = 1; sign_link = line->ops->sign_link_up(&unit_data, line, @@ -309,6 +310,13 @@ static int ipaccess_rcvmsg(struct e1inp_line *line, struct msgb *msg, "could not register FD\n"); goto err; } + + /* set TCP_NODELAY (FIXME: just call update_fd_settings() here instead?) */ + ret = setsockopt(newbfd->fd, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); + if (ret < 0) + LOGP(DLINP, LOGL_ERROR, "Failed to set TCP_NODELAY: %s\n", strerror(errno)); + + /* now we can release the dummy RSL line. */ e1inp_line_put(line); @@ -587,6 +595,11 @@ static void update_fd_settings(struct e1inp_line *line, int fd) #endif #endif } + + val = 1; + ret = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &val, sizeof(val)); + if (ret < 0) + LOGP(DLINP, LOGL_ERROR, "Failed to set TCP_NODELAY: %s\n", strerror(errno)); }
Actions