Skip to content

N
net
Switch branch/tag
  1. 21 Jun, 2017 3 commits
  3. 19 Jun, 2017 2 commits
  5. 16 Jun, 2017 1 commit
  7. 14 Jun, 2017 2 commits
  9. 13 Jun, 2017 2 commits
  11. 10 Jun, 2017 1 commit
  13. 09 Jun, 2017 2 commits
  15. 05 Jun, 2017 2 commits
  17. 02 Jun, 2017 1 commit
  19. 29 May, 2017 1 commit
  21. 27 May, 2017 2 commits
    • Mikio Hara's avatar
      ipv6: add {Read,Write}Batch methods to PacketConn · 48359f4f
      Mikio Hara authored 
      This change provides message IO functionality that may support the
construction of modern datagram transport protocols.

The modern datagram transport protocols on a multihomed node basically
need to control each packet path for traffic engineering by using
information belongs to network- or link-layer implementation. In
addtion, it's desirable to be able to do simultaneous transmission
across multiple network- or link-layer adjacencies wihtout any
additional cost.

The ReadBatch and WriteBatch methods of PacketConn can be used to read
and write an IO message that contains the information of network- or
link-layer implementation, and read and write a batch of IO messages
on Linux. The Marshal and Parse methods of ControlMessage can help to
marshal and parse information contained in IO messages.

Updates golang/go#3661.

Change-Id: I94484e2e135f4969ddf5c2548bf6be0cc16888bf
Reviewed-on: https://go-review.googlesource.com/38276
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarIan Lance Taylor <iant@golang.org>
      48359f4f
    • Mikio Hara's avatar
      ipv4: add {Read,Write}Batch methods to {Packet,Raw}Conn · b7a1f62a
      Mikio Hara authored 
      This change provides message IO functionality that may support the
construction of modern datagram transport protocols.

The modern datagram transport protocols on a multihomed node basically
need to control each packet path for traffic engineering by using
information belongs to network- or link-layer implementation. In
addtion, it's desirable to be able to do simultaneous transmission
across multiple network- or link-layer adjacencies wihtout any
additional cost.

The ReadBatch and WriteBatch methods of PacketConn and RawConn can be
used to read and write an IO message that contains the information of
network- or link-layer implementation, and read and write a batch of
IO messages on Linux. The Marshal and Parse methods of ControlMessage
and Header can help to marshal and parse information contained in IO
messages.

Updates golang/go#3661.

Change-Id: Ia84a9d3bc51641406eaaf4258f2a3066945cc323
Reviewed-on: https://go-review.googlesource.com/38275
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarIan Lance Taylor <iant@golang.org>
      b7a1f62a
  23. 26 May, 2017 4 commits
    • Tom Bergan's avatar
      http2: fix nil dereference after Read completes with an error · 3470a06c
      Tom Bergan authored 
      Case happens if Read is called after it has already returned an error
previously. Verified that the new TestPipeCloseWithError test fails
before this change but passes after.

Updates golang/go#20501

Change-Id: I636fbb194f2d0019b0722556cc25a88da2d18e13
Reviewed-on: https://go-review.googlesource.com/44330
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarBrad Fitzpatrick <bradfitz@golang.org>
      3470a06c
    • David Lazar's avatar
      http2: fix typo in comment · 06b2bf2f
      David Lazar authored 
      Change-Id: I0febb967cf55118693952332bbb6f740616cfd39
Reviewed-on: https://go-review.googlesource.com/44271
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarBrad Fitzpatrick <bradfitz@golang.org>
      06b2bf2f
    • Mikio Hara's avatar
      internal/socket: add message IO functionality · b8b13433
      Mikio Hara authored 
      This change adds portable message IO methods of Conn, IO message and
control message types, and parse methods for the types to provide
those functionality to the ipv4 and ipv6 packages.

With this change, the ipv4 and ipv6 packages can provide low-level
(but less heap allocation and the cost of invoking system calls) API
such as read and write operations for a batch of IO messages.

On vanilla linux/amd64 virtual machine:
BenchmarkUDP/Iter-1-2            1000000              8068 ns/op             408 B/op         14 allocs/op
BenchmarkUDP/Batch-1-2           1000000              8610 ns/op             440 B/op         14 allocs/op
BenchmarkUDP/Iter-2-2             500000             15390 ns/op             816 B/op         28 allocs/op
BenchmarkUDP/Batch-2-2            500000             12715 ns/op             696 B/op         20 allocs/op
BenchmarkUDP/Iter-4-2             200000             30763 ns/op            1632 B/op         56 allocs/op
BenchmarkUDP/Batch-4-2            300000             21853 ns/op            1216 B/op         32 allocs/op
BenchmarkUDP/Iter-8-2             100000             61460 ns/op            3264 B/op        112 allocs/op
BenchmarkUDP/Batch-8-2            200000             39048 ns/op            2256 B/op         56 allocs/op
BenchmarkUDP/Iter-16-2             50000            122408 ns/op            6528 B/op        224 allocs/op
BenchmarkUDP/Batch-16-2           100000             72728 ns/op            4336 B/op        104 allocs/op
BenchmarkUDP/Iter-32-2             30000            243137 ns/op           13056 B/op        448 allocs/op
BenchmarkUDP/Batch-32-2            50000            141332 ns/op            8496 B/op        200 allocs/op
BenchmarkUDP/Iter-64-2             20000            488125 ns/op           26112 B/op        896 allocs/op
BenchmarkUDP/Batch-64-2            30000            282078 ns/op           16816 B/op        392 allocs/op
BenchmarkUDP/Iter-128-2            10000            973752 ns/op           52224 B/op       1792 allocs/op
BenchmarkUDP/Batch-128-2           10000            551021 ns/op           33456 B/op        776 allocs/op
BenchmarkUDP/Iter-256-2             3000           1977852 ns/op          104448 B/op       3584 allocs/op
BenchmarkUDP/Batch-256-2           10000           1252596 ns/op           66736 B/op       1544 allocs/op
BenchmarkUDP/Iter-512-2             2000           4147495 ns/op          208896 B/op       7168 allocs/op
BenchmarkUDP/Batch-512-2            3000           2175774 ns/op          121128 B/op       2612 allocs/op

Change-Id: I3e08b28917b62dbea7936a86acb24e25ccaf5365
Reviewed-on: https://go-review.googlesource.com/38212
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarIan Lance Taylor <iant@golang.org>
      b8b13433
    • Mikio Hara's avatar
      internal/socket: add platform-dependent boilerplate files · f61a7739
      Mikio Hara authored 
      Change-Id: I951a1abc36083401d183d61c58a8186b5ce41c6c
Reviewed-on: https://go-review.googlesource.com/38211Reviewed-by: 's avatarIan Lance Taylor <iant@golang.org>
      f61a7739
  25. 25 May, 2017 1 commit
  27. 24 May, 2017 9 commits
  29. 23 May, 2017 1 commit
  31. 22 May, 2017 3 commits
  33. 17 May, 2017 1 commit
  35. 15 May, 2017 1 commit
    • Brad Fitzpatrick's avatar
      http2: reduce the number of select cases in serverConn.server · 34057069
      Brad Fitzpatrick authored 
      This drops the number of select cases in serverConn.server from 10 to
6. It was 7 in Go 1.7. It increased to 10 in Go 1.8.

* replace testing-only testHookCh with always-on grab-bag serveMsgCh
  to be used for both test purposes, and infrequently used message
  types

* remove the settingsTimer.C case for the initial settings timer
  and use serveMsgCh and time.AfterFunc instead

* ... and do the same for the idle timeout timer.

* ... and for the shutdown timer.

* remove wantStartPushCh and just send the *startPushRequest to
  serveMsgCh too

I could go further with this (and plan to, later), but these are the
safe and easy ones that don't require more work elsewhere.

The speed gets better the more the request/response go via the
serverConn.serve loop. (once per Request.Body.Read or
ResponseWriter.Flush):

  name            old time/op    new time/op  delta
  ServerGets-4    138µs ± 3%     134µs ± 2%   -2.54%   (p=0.002 n=10+10)
  ServerPosts-4   176µs ±27%     154µs ± 2%   -12.62%  (p=0.011 n=10+10)

Updates kubernetes/kubernetes#45216
Updates golang/go#20302

Change-Id: I18019554089d7e3d76355d7137b5957e9597e803
Reviewed-on: https://go-review.googlesource.com/43034
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarTom Bergan <tombergan@google.com>
      34057069
  37. 13 May, 2017 1 commit
    • Tom Bergan's avatar
      http2: fix lock contention slowdown due to gracefulShutdownCh · 84f0e6f9
      Tom Bergan authored 
      gracefulShutdownCh is shared by all connections in a server. When a
server accumulates many connections (e.g., 5000 in the kubemark-5000
benchmark), we have 5000 serverConn.serve goroutines selecting on this
channel. This means 5000 goroutines hammer the channel's lock, which
causes severe lock contention.

The fix in this CL is to make a local proxy for gracefulShutdownCh in
each connection so that each connection selects on gracefulShutdownCh
at most once per connection rather than once per serverConn.serve loop
iteration.

This fix is intended to be backported quickly into Go 1.8.2. The
downside of this fix is 2KB extra stack usage per connection. A better
fix will be implemented in Go 1.9.

Unfortunately, I have been unable to reproduce this problem locally.
This fix was verified by the kubernetes team. See:
https://github.com/kubernetes/kubernetes/issues/45216#issuecomment-300830243

Updates golang/go#20302

Change-Id: I19ab19268a6ccab9b6e9dffa0cfbc89b8c7d0f19
Reviewed-on: https://go-review.googlesource.com/43455
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarBrad Fitzpatrick <bradfitz@golang.org>
      84f0e6f9