netstat display Huge amount of "SYN_RCVD" state - performance

Question

netstat display Huge amount of "SYN_RCVD" state on Solaris 11, it's a nginx reverse server.
# netstat -na | awk '/^20/ {++S[$NF]} END {for(a in S) print a, S[a]}'
TIME_WAIT 515
ESTABLISHED 427
SYN_SENT 14
LAST_ACK 223
Connected 9488
FIN_WAIT_1 37
FIN_WAIT_2 167
CLOSING 48
CLOSE_WAIT 11
Idle 5
SYN_RCVD 4437
The server tcp/ip parameters:
# ndd -get /dev/tcp tcp_time_wait_interval
60000
# ndd -get /dev/tcp tcp_keepalive_interval
15000
# ndd -get /dev/tcp tcp_fin_wait_2_flush_interval
67500
# ndd -get /dev/tcp tcp_conn_req_max_q
16384
# ndd -get /dev/tcp tcp_conn_req_max_q0
16384
# ndd -get /dev/tcp tcp_xmit_hiwat
400000
# ndd -get /dev/tcp tcp_recv_hiwat
400000
# ndd -get /dev/tcp tcp_cwnd_max
2097152
# ndd -get /dev/tcp tcp_ip_abort_interval
20000
# ndd -get /dev/tcp tcp_rexmit_interval_initial
4000
# ndd -get /dev/tcp tcp_rexmit_interval_max
10000
# ndd -get /dev/tcp tcp_rexmit_interval_min
3000
# ndd -get /dev/tcp tcp_max_buf
4194304
How to tuning this nginx reverse server.
Thanks.

Answer 1

# of conn reqs in SYN_RCVD - "fully-established" connections -
those which have finished the 3-way handshake and
are waiting to be picked up by an "accept()" call.
You should consider tuning your accept_mutex nginx parameter and may be try different nginx processing method, just to get if it is real reason.