[Cake] a couple quick test runs of the v17+1 patchset

[Cake] a couple quick test runs of the v17+1 patchset

[Dave Taht]

The paper makes the claim that we can shape to .1% of the bandwidth
and win. Well, at gigE rates, it's closer to .2%, as setting the
shaper to 980mbit got consistently sub 2.2ms RTT latency on the rrul
test, higher usually cracked 5 or 6, while bandwidth went from 934 to
936. So we can take a .2%-ish bandwidth hit in exchange for reducing
the max latency by 2/3s on this particular test series.

And this of course ignores the impact of flow control, my environment,
academic rigor, etc, etc. And I shudder to go measure the cpu
overhead.

And it didn't crash, and perhaps I'll go play with cake with veth
simulating > gigE bandwidths.

Only question is: why is the capacity estimate 970mbit?

root@spaceheater:~/git/linux/tc-adv/tc# ./tc -s qdisc show dev enp6s0
qdisc cake 8017: root refcnt 2 bandwidth 980Mbit diffserv3
triple-isolate split-gso rtt 100.0ms raw overhead 0
 Sent 23871634960 bytes 17506050 pkt (dropped 2, overlimits 30089616 requeues 7)
 backlog 370996b 246p requeues 7
 memory used: 1053008b of 15140Kb
 capacity estimate: 970Mbit
 min/max network layer size:           42 /    1514
 min/max overhead-adjusted size:       42 /    1514
 average network hdr offset:           14

                   Bulk  Best Effort        Voice
  thresh      61250Kbit      980Mbit      245Mbit
  target          5.0ms        5.0ms        5.0ms
  interval      100.0ms      100.0ms      100.0ms
  pk_delay        4.1ms        3.2ms        4.2ms
  av_delay        2.3ms        2.4ms        3.1ms
  sp_delay        231us        642us        321us
  backlog        27318b      263436b       80242b
  pkts          1476388     11611873      4418037
  bytes      1497338404  16405801643   5968868937
  way_inds            0            8            0
  way_miss           12          209           15
  way_cols            0            0            0
  drops               2            0            0
  marks               0            0            0
  ack_drop            0            0            0
  sp_flows            2            4            3
  bk_flows            1            2            1
  un_flows            0            0            0
  max_len         68130        68130        68130
  quantum          1514         1514         1514






-- 

Dave Täht
CEO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-669-226-2619

Re: [Cake] a couple quick test runs of the v17+1 patchset

[Dave Taht]

And I was being an idiot. Using the ethernet keyword to account for
the framing overhead AND shaping to 1Gbit,
I got local latency results twice as good as what I can get from
sch_fq, no muss, no fuss, and no loss of bandwidth vs the unshaped
mode.

Hardware flow control will still defeat us, but....

Summary of rrul test run 'v17-cake-1gbit-ethernet' (at 2018-05-30
20:48:14.853980):

                             avg       median          # data pts
 Ping (ms) ICMP   :         1.74         1.88 ms              347
 Ping (ms) UDP BE :       879.31         2.21 ms              349
 Ping (ms) UDP BK :       853.93         2.25 ms              349
 Ping (ms) UDP EF :       879.93         2.21 ms              349
 Ping (ms) avg    :       653.73         2.13 ms              350
 TCP download BE  :       230.61       230.58 Mbits/s         299
 TCP download BK  :       230.81       230.70 Mbits/s         299
 TCP download CS5 :       230.61       230.66 Mbits/s         299
 TCP download EF  :       230.21       230.58 Mbits/s         299
 TCP download avg :       230.56       230.62 Mbits/s         300
 TCP download sum :       922.24       922.48 Mbits/s         300
 TCP totals       :      1858.68      1859.31 Mbits/s         300
 TCP upload BE    :       322.47       322.56 Mbits/s         290
 TCP upload BK    :        57.57        57.47 Mbits/s         271
 TCP upload CS5   :       322.39       322.49 Mbits/s         276
 TCP upload EF    :       234.01       234.18 Mbits/s         281
 TCP upload avg   :       234.11       234.18 Mbits/s         300
 TCP upload sum   :       936.44       936.72 Mbits/s         300

but the capacity estimate is still 970?

qdisc cake 8017: root refcnt 2 bandwidth 1Gbit diffserv3
triple-isolate split-gso rtt 100.0ms noatm overhead 38 mpu 84
 Sent 51504294511 bytes 37724591 pkt (dropped 6, overlimits 64958695
requeues 12)
 backlog 0b 0p requeues 12
 memory used: 1053008b of 15140Kb
 capacity estimate: 970Mbit
 min/max network layer size:           28 /    1500
 min/max overhead-adjusted size:       84 /    1538
 average network hdr offset:           14

                   Bulk  Best Effort        Voice
  thresh      62500Kbit        1Gbit      250Mbit
  target          5.0ms        5.0ms        5.0ms
  interval      100.0ms      100.0ms      100.0ms
  pk_delay          5us          5us          6us
  av_delay          3us          2us          2us
  sp_delay          2us          1us          1us
  backlog            0b           0b           0b
  pkts          3164050     25030267      9530280
  bytes      3227519915  35396974782  12879808898
  way_inds            0            8            0
  way_miss           21          366           25
  way_cols            0            0            0
  drops               5            0            1
  marks               0            0            0
  ack_drop            0            0            0
  sp_flows            1            3            0
  bk_flows            0            1            1
  un_flows            0            0            0
  max_len         68130        68130        68130