NSPolicy #64

kedixa · 2025-05-17T12:08:11Z

No description provided.

To support better address break strategy.

When tested on Github macOS, this difference will exceed expectations, resulting in false test failures.

github-actions · 2025-05-17T12:13:00Z

Benchmark

The benchmark runs on Github Actions.

Environment

Details

Architecture:                         x86_64
CPU op-mode(s):                       32-bit, 64-bit
Address sizes:                        48 bits physical, 48 bits virtual
Byte Order:                           Little Endian
CPU(s):                               4
On-line CPU(s) list:                  0-3
Vendor ID:                            AuthenticAMD
Model name:                           AMD EPYC 7763 64-Core Processor
CPU family:                           25
Model:                                1
Thread(s) per core:                   2
Core(s) per socket:                   2
Socket(s):                            1
Stepping:                             1
BogoMIPS:                             4890.85
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl tsc_reliable nonstop_tsc cpuid extd_apicid aperfmperf tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy svm cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw topoext vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves user_shstk clzero xsaveerptr rdpru arat npt nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold v_vmsave_vmload umip vaes vpclmulqdq rdpid fsrm
Virtualization:                       AMD-V
Hypervisor vendor:                    Microsoft
Virtualization type:                  full
L1d cache:                            64 KiB (2 instances)
L1i cache:                            64 KiB (2 instances)
L2 cache:                             1 MiB (2 instances)
L3 cache:                             32 MiB (1 instance)
NUMA node(s):                         1
NUMA node0 CPU(s):                    0-3
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Not affected
Vulnerability Spec rstack overflow:   Vulnerable: Safe RET, no microcode
Vulnerability Spec store bypass:      Vulnerable
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Retpolines; STIBP disabled; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

RANGE                                 SIZE  STATE REMOVABLE  BLOCK
0x0000000000000000-0x000000003fffffff   1G online       yes    0-7
0x0000000100000000-0x00000004bfffffff  15G online       yes 32-151

Memory block size:       128M
Total online memory:      16G
Total offline memory:      0B

Go

Bench command bench_go --concurrency 64 --compute 8 --max-secs 60 --total 200000 --times 5 -y

Details

name	cost	times	mean(ms)	stddev	per sec
wf_go_one_name	755	5	151.00	71.44	1324503
wf_go_five_name	393	5	78.60	3.36	2544529
wf_go_ten_name	339	5	67.80	0.45	2949852

go_one_name	1149	5	229.80	24.61	870322
go_five_name	412	5	82.40	1.52	2427184
go_ten_name	347	5	69.40	0.89	2881844

switch_one_name	1174	5	234.80	27.34	851788
switch_five_name	401	5	80.20	2.86	2493765
switch_ten_name	350	5	70.00	1.00	2857142

Timer

Bench command bench_timer --concurrency 4096 --handler 12 --poller 6 --compute 8 --max-secs 60 --total 200000 --times 5 -y

Details

name	cost	times	mean(ms)	stddev	per sec
wf_repeat	660	5	132.00	3.39	1515151
default_timer	647	5	129.40	3.21	1545595
yield	390	5	78.00	1.87	2564102
timer_in_task	645	5	129.00	4.06	1550387

timer_by_name	1851	5	370.20	3.90	540248
cancel_by_name	797	5	159.40	13.30	1254705
detach_by_name	1695	5	339.00	2.83	589970
detach3_by_name	4945	5	989.00	7.38	202224
one_name	815	5	163.00	3.54	1226993
two_name	841	5	168.20	6.98	1189060
ten_name	889	5	177.80	4.38	1124859
name_one_by_one	1168	5	233.60	17.53	856164

timer_by_id	854	5	170.80	3.03	1170960
timer_by_addr	920	5	184.00	2.55	1086956
cancel_by_id	678	5	135.60	2.61	1474926
detach_by_id	1005	5	201.00	3.67	995024
detach_inf_by_id	922	5	184.40	1.67	1084598
one_id	838	5	167.60	4.04	1193317
two_id	727	5	145.40	2.07	1375515
ten_id	643	5	128.60	2.30	1555209
id_one_by_one	1570	5	314.00	13.45	636942

Dag

Bench command bench_graph --num-nodes 128 --total 500 --group-size 10 --task-per-node 3 --max-secs 60 --times 5 --handler 12 --poller 6 -y

Details

name	cost	times	mean(ms)	stddev	per sec
wf_chain	870	5	174.00	4.53	2873
coke_chain_once	863	5	172.60	4.04	2896
coke_chain	825	5	165.00	4.18	3030

wf_tree	1067	5	213.40	2.61	2343
coke_tree_once	1123	5	224.60	2.70	2226
coke_tree	1064	5	212.80	0.84	2349

wf_net	1132	5	226.40	2.97	2208
coke_net_once	1169	5	233.80	2.59	2138
coke_net	1066	5	213.20	1.10	2345

wf_flower	943	5	188.60	0.89	2651
coke_flower_once	964	5	192.80	1.64	2593
coke_flower	928	5	185.60	1.82	2693

Exception

Bench command bench_exception --concurrency 4096 --total 200000 --max-secs 60 --times 5 --handler 12 --poller 6 -y

Details

name	cost	times	mean(ms)	stddev	per sec
normal_yield	438	5	87.60	1.67	2283105
yield_catch	434	5	86.80	1.48	2304147

d1_p0	553	5	110.60	0.89	1808318
d2_p0	552	5	110.40	2.51	1811594
d5_p0	582	5	116.40	2.51	1718213

d1_p100	1724	5	344.80	7.73	580046
d2_p100	2507	5	501.40	2.70	398883
d5_p100	4852	5	970.40	4.22	206100

d1_p1	535	5	107.00	2.24	1869158
d1_p5	596	5	119.20	0.45	1677852
d1_p10	636	5	127.20	0.84	1572327
d1_p20	702	5	140.40	1.52	1424501
d1_p50	1140	5	228.00	1.00	877192

Queue

Bench command bench_queue --concurrency 10 --total 1000000 --max-secs 60 --times 5 --que-size 1000 --batch-size 10 --handler 12 --poller 6 -y

Details

name	cost	times	mean(ms)	stddev	per sec
try_push_pop	1293	5	258.60	7.92	3866976
push_pop	1768	5	353.60	5.77	2828054
push_pop_range	318	5	63.60	14.36	15723270

Weighted NSPolicy

Bench command ./bazel-bin/benchmark/bench_weighted_policy --concurrency 1 --total 1000000 --times 5 -y

Details

name	num addrs	fail ratio	cost(ms)	times	mean(ms)	stddev	per sec
weighted_random	1000	1/10000	244	5	48.80	0.45	20491803
weighted_random	10000	1/10000	399	5	79.80	0.45	12531328
weighted_random	100000	1/10000	879	5	175.80	8.84	5688282
weighted_random	1000000	1/10000	1916	5	383.20	41.69	2609603

weighted_random	10000	1/1000	402	5	80.40	0.55	12437810
weighted_random	10000	1/100	412	5	82.40	0.55	12135922
weighted_random	10000	1/10	515	5	103.00	0.71	9708737

weighted_least_conn	1000	1/10000	497	5	99.40	0.55	10060362
weighted_least_conn	10000	1/10000	738	5	147.60	0.55	6775067
weighted_least_conn	100000	1/10000	1202	5	240.40	1.14	4159733
weighted_least_conn	1000000	1/10000	1472	5	294.40	0.55	3396739

weighted_least_conn	10000	1/1000	737	5	147.40	0.55	6784260
weighted_least_conn	10000	1/100	745	5	149.00	0.00	6711409
weighted_least_conn	10000	1/10	821	5	164.20	0.45	6090133

weighted_round_robin	1000	1/10000	354	5	70.80	0.84	14124293
weighted_round_robin	10000	1/10000	453	5	90.60	0.55	11037527
weighted_round_robin	100000	1/10000	570	5	114.00	2.92	8771929
weighted_round_robin	1000000	1/10000	1035	5	207.00	6.60	4830917

weighted_round_robin	10000	1/1000	455	5	91.00	0.71	10989010
weighted_round_robin	10000	1/100	482	5	96.40	0.55	10373443
weighted_round_robin	10000	1/10	647	5	129.40	0.55	7727975

kedixa · 2025-05-17T13:45:49Z

weighted_random策略在地址数量较少时效率最优，但随着地址数量增加，效率下降的速度也最快，这是因为其随机性导致几乎次次无法命中高速缓存，幸好实际业务不需要同时为百万个地址做负载均衡。

kedixa added 29 commits April 4, 2025 21:49

feature: coke::AddressInfo

6447619

feature: coke::NSPolicy

0e6d0ed

feature: coke::NSPolicy impl

a83199e

format code style

41701af

feature: coke::WeightedPolicyBase

02165b4

feature: coke::WeightedPolicyBase impl

fb492a1

optimize: detail::RefCounted ref count type use uint32_t

74a2d53

format code style

f02ac69

feature: coke::detail::BinaryIndexedTree

fc5f482

fix: include guard for weighted_policy_base_impl.h

12d69a1

feature: coke::WeightedRandomPolicy

4991299

feature: coke::WeightedRandomPolicy impl

86b07df

build: nspolicy

6f25eab

feature: coke::WeightedLeastConnPolicy

ceef365

Using fail_marks instead of fail_cnt

416e040

To support better address break strategy.

feature: coke::WeightedLeastConnPolicy impl

f3bbd3f

feature: coke::WeightedRoundRobinPolicy

b083285

WeightedPolicyBase use NoNeedFinishTag instead of NeedFinishTag

ef9a90a

test: nspolicy

86e256d

benchmark: bench_weighted_policy

4b76560

add bench_weighted_policy to workflows

8743541

fix: clang 15 not support this aggregate initialization

ea0b029

docs: nspolicy

9f7580b

build: disable some warnings

8daf8fa

Merge branch 'master' into dev

4c90e92

fix: add override

2159196

test: remove EXPECT_NEAR check in test_sleep

78e645b

When tested on Github macOS, this difference will exceed expectations, resulting in false test failures.

ci: add ci on macos

dd0654f

version: v0.5.0

12a8064

kedixa added the do-benchmark Let github action perform benchmarking label May 17, 2025

github-actions bot removed the do-benchmark Let github action perform benchmarking label May 17, 2025

kedixa merged commit ddb92c9 into master May 17, 2025
24 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

NSPolicy #64

NSPolicy #64

Uh oh!

kedixa commented May 17, 2025

Uh oh!

github-actions bot commented May 17, 2025

Uh oh!

kedixa commented May 17, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

NSPolicy #64

NSPolicy #64

Uh oh!

Conversation

kedixa commented May 17, 2025

Uh oh!

github-actions bot commented May 17, 2025

Benchmark

Environment

Go

Timer

Dag

Exception

Queue

Weighted NSPolicy

Uh oh!

kedixa commented May 17, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants