You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Linux 3.13 enables TCP fast open, RFC 7413, by default, so the feature also must be supported by Tempesta for client and server sockets:
Synchronous sockets must be adjusted to handle TFO (like setsockopt(TCP_FASTOPEN) for server sockets and sendto(MSG_FASTOPEN), if an HTTP request of TLS record (see Full TLS proxying #769) to be sent fit MSS, for client sockets). The queue limit size shall be configurable and 1000 by default.
duplicated SYN+data segments must be handled in an idempotent fashion, i.e. we must not pass nonidepotent requests for TFO (in sense of Redesign of TCP synchronous sending and data caching #391). It seems there is nothing to do for this point for client connections.
tempesta.sh must ensure that tcp_tw_reuse isn't used together with tcp_fastopen to avoid duplicate data SYNs (see chapter 3.5 in the original paper.
Linux 3.13 enables TCP fast open, RFC 7413, by default, so the feature also must be supported by Tempesta for client and server sockets:
setsockopt(TCP_FASTOPEN)
for server sockets andsendto(MSG_FASTOPEN)
, if an HTTP request of TLS record (see Full TLS proxying #769) to be sent fit MSS, for client sockets). The queue limit size shall be configurable and1000
by default.tempesta.sh
must ensure thattcp_tw_reuse
isn't used together withtcp_fastopen
to avoid duplicate data SYNs (see chapter 3.5 in the original paper.See TCP Fast Open: expediting web services for details. Ideal HTTP Performance refers the feature as crucial for HTTP performance.
The text was updated successfully, but these errors were encountered: