linux_dsm_epyc7002/net/dccp
Eric Dumazet 789f558cfb tcp/dccp: get rid of central timewait timer
Using a timer wheel for timewait sockets was nice ~15 years ago when
memory was expensive and machines had a single processor.

This does not scale, code is ugly and source of huge latencies
(Typically 30 ms have been seen, cpus spinning on death_lock spinlock.)

We can afford to use an extra 64 bytes per timewait sock and spread
timewait load to all cpus to have better behavior.

Tested:

On following test, /proc/sys/net/ipv4/tcp_tw_recycle is set to 1
on the target (lpaa24)

Before patch :

lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
419594

lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
437171

While test is running, we can observe 25 or even 33 ms latencies.

lpaa24:~# ping -c 1000 -i 0.02 -qn lpaa23
...
1000 packets transmitted, 1000 received, 0% packet loss, time 20601ms
rtt min/avg/max/mdev = 0.020/0.217/25.771/1.535 ms, pipe 2

lpaa24:~# ping -c 1000 -i 0.02 -qn lpaa23
...
1000 packets transmitted, 1000 received, 0% packet loss, time 20702ms
rtt min/avg/max/mdev = 0.019/0.183/33.761/1.441 ms, pipe 2

After patch :

About 90% increase of throughput :

lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
810442

lpaa23:~# ./super_netperf 200 -H lpaa24 -t TCP_CC -l 60 -- -p0,0
800992

And latencies are kept to minimal values during this load, even
if network utilization is 90% higher :

lpaa24:~# ping -c 1000 -i 0.02 -qn lpaa23
...
1000 packets transmitted, 1000 received, 0% packet loss, time 19991ms
rtt min/avg/max/mdev = 0.023/0.064/0.360/0.042 ms

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-04-13 16:40:05 -04:00
..
ccids dccp: re-enable debug macro 2014-02-16 23:45:00 -05:00
ackvec.c dccp: replace min/casting by min_t 2014-11-18 15:26:32 -05:00
ackvec.h net: dccp: Remove extern from function prototypes 2013-10-19 19:12:11 -04:00
ccid.c net/dccp/ccid.c: add __init to ccid_activate 2014-10-01 18:33:13 -04:00
ccid.h net: dccp: Remove extern from function prototypes 2013-10-19 19:12:11 -04:00
dccp.h ipv4: dccp: handle ICMP messages on DCCP_NEW_SYN_RECV request sockets 2015-03-23 16:52:26 -04:00
diag.c inet_diag: add const to inet_diag_req_v2 2015-03-10 13:45:28 -04:00
feat.c dccp: kerneldoc warning fixes 2014-11-18 15:26:31 -05:00
feat.h net: dccp: Remove extern from function prototypes 2013-10-19 19:12:11 -04:00
input.c dccp: spelling s/reseting/resetting 2014-11-18 15:26:32 -05:00
ipv4.c inet: fix double request socket freeing 2015-03-23 21:40:48 -04:00
ipv6.c ipv6: dccp: handle ICMP messages on DCCP_NEW_SYN_RECV request sockets 2015-03-23 16:52:26 -04:00
ipv6.h inet: includes a sock_common in request_sock 2013-10-10 00:08:07 -04:00
Kconfig net/dccp: remove depends on CONFIG_EXPERIMENTAL 2013-01-11 11:39:34 -08:00
Makefile dccp: Policy-based packet dequeueing infrastructure 2010-12-07 13:47:12 +01:00
minisocks.c tcp/dccp: get rid of central timewait timer 2015-04-13 16:40:05 -04:00
options.c dccp: remove obsolete code 2014-01-04 20:18:49 -05:00
output.c ipv4: add a sock pointer to ip_queue_xmit() 2014-04-15 12:58:34 -04:00
probe.c net: Remove iocb argument from sendmsg and recvmsg 2015-03-02 13:06:31 -05:00
proto.c net: Remove iocb argument from sendmsg and recvmsg 2015-03-02 13:06:31 -05:00
qpolicy.c dccp qpolicy: Parameter checking of cmsg qpolicy parameters 2010-12-07 13:47:12 +01:00
sysctl.c dccp: make the request_retries minimum is 1 2014-05-14 15:34:16 -04:00
timer.c inet: get rid of central tcp/dccp listener timer 2015-03-20 12:40:25 -04:00