Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Refactor] Asynchronous llumlet #56

Merged
merged 6 commits into from
Oct 16, 2024
Merged

[Refactor] Asynchronous llumlet #56

merged 6 commits into from
Oct 16, 2024

Conversation

ZeldaHuang
Copy link
Contributor

  1. use async api in vllm backend.
  2. make llumlet async actor .

Copy link

prefill p25 p50 p75 p95 p99 mean
latency(ms) 34378.30 92614.34 213082.56 241795.76 247962.89 112179.54
decode p25 p50 p75 p95 p99 mean
latency(ms) 51.91 57.36 68.75 107.71 233.96 66.89

Copy link

migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 192.00 MB 304.00 MB 312.00 MB 328.00 MB 352.00 MB 360.00 MB 472.00 MB 544.00 MB 560.00 MB 600.00 MB
rpc_speed(GB/s) 1.03 1.54 1.76 1.90 2.01 2.12 2.15 2.15 2.20 2.32 2.37 2.35 2.38 2.41 2.40 2.52 2.57 2.40 2.44 2.39 2.66 2.58 2.80 2.72 2.71 2.75 2.23 2.69 3.10 3.19 2.97 3.48
migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 184.00 MB 192.00 MB 200.00 MB 312.00 MB 472.00 MB 480.00 MB 552.00 MB 560.00 MB 568.00 MB 696.00 MB 728.00 MB
gloo_speed(GB/s) 1.05 1.64 2.12 2.46 2.56 2.64 2.80 3.10 2.70 3.14 3.19 2.76 3.35 3.74 3.01 3.01 2.75 2.86 2.83 2.53 3.00 2.64 2.73 3.35 3.14 2.01 3.20 3.05 3.26 2.97 2.73 3.20
migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 312.00 MB 320.00 MB 328.00 MB 480.00 MB 512.00 MB 536.00 MB 728.00 MB
nccl_speed(GB/s) 0.19 0.44 0.67 0.81 0.89 1.19 1.40 1.49 1.51 2.12 2.10 2.13 1.80 2.39 2.65 2.53 2.77 3.00 2.98 2.81 3.23 3.62 3.89 3.82 4.17 4.73 6.13 4.25 5.94

Copy link

prefill p25 p50 p75 p95 p99 mean
latency(ms) 35827.43 95055.36 201583.15 241373.11 271782.55 115871.22
decode p25 p50 p75 p95 p99 mean
latency(ms) 53.70 58.26 70.75 126.07 253.15 72.26

Copy link

prefill p25 p50 p75 p95 p99 mean
latency(ms) 27593.03 105515.88 193648.63 235307.06 244263.45 114101.05
decode p25 p50 p75 p95 p99 mean
latency(ms) 53.30 58.20 68.23 106.19 210.36 66.29

Copy link

migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 176.00 MB 200.00 MB 208.00 MB 280.00 MB 312.00 MB 472.00 MB 552.00 MB
rpc_speed(GB/s) 1.03 1.51 1.81 1.89 2.01 2.12 2.13 2.19 2.22 2.33 2.36 2.34 2.38 2.41 2.39 2.47 2.45 2.58 2.42 2.55 2.55 2.52 2.48 2.58 2.75 3.48 3.40
migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 192.00 MB 200.00 MB 280.00 MB 296.00 MB 304.00 MB 312.00 MB 472.00 MB 488.00 MB 536.00 MB
gloo_speed(GB/s) 1.06 1.72 2.07 2.48 2.69 2.65 2.75 2.99 2.92 3.07 3.45 3.36 3.60 3.63 3.61 2.98 2.62 3.21 3.42 3.00 1.32 3.28 2.83 2.44 0.69 3.48 3.01 2.70 2.25 2.93 2.93
migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 192.00 MB 312.00 MB 320.00 MB 328.00 MB 480.00 MB 512.00 MB 536.00 MB 552.00 MB 576.00 MB 584.00 MB 696.00 MB 960.00 MB
nccl_speed(GB/s) 0.18 0.44 0.59 0.83 0.88 1.27 1.48 1.51 1.42 1.67 2.04 1.96 2.42 2.31 3.04 2.23 3.21 3.56 2.93 3.03 3.23 2.80 3.20 2.66 4.07 4.50 4.95 5.98 4.27 3.11 4.70 4.54 4.85 5.98

Copy link

prefill p25 p50 p75 p95 p99 mean
latency(ms) 31760.35 103447.93 199532.66 305825.09 308272.81 122931.24
decode p25 p50 p75 p95 p99 mean
latency(ms) 52.94 56.95 65.23 102.87 206.59 65.60

Copy link

prefill p25 p50 p75 p95 p99 mean
latency(ms) 24244.81 117511.61 204478.41 260129.75 280649.62 121291.87
decode p25 p50 p75 p95 p99 mean
latency(ms) 53.24 57.96 68.85 115.24 197.89 67.06

Copy link

migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 192.00 MB 280.00 MB 312.00 MB 320.00 MB 360.00 MB 472.00 MB 480.00 MB 552.00 MB
rpc_speed(GB/s) 1.06 1.55 1.82 1.93 2.06 2.10 2.18 2.20 2.22 2.33 2.52 2.36 2.43 2.32 2.35 2.52 2.54 2.51 2.44 2.56 2.57 2.66 2.88 2.25 2.65 2.71 2.84 2.99 3.18 3.27
migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 192.00 MB 312.00 MB 320.00 MB 344.00 MB 472.00 MB 592.00 MB
gloo_speed(GB/s) 1.04 1.62 2.08 2.26 2.57 2.65 2.99 2.99 2.76 3.50 3.09 3.34 3.40 3.11 2.80 2.54 2.84 2.58 2.42 2.62 2.74 3.41 3.56 2.71 3.44 2.92 2.79 3.35
migration_size 8.00 MB 16.00 MB 24.00 MB 32.00 MB 40.00 MB 48.00 MB 56.00 MB 64.00 MB 72.00 MB 80.00 MB 88.00 MB 96.00 MB 104.00 MB 112.00 MB 120.00 MB 128.00 MB 136.00 MB 144.00 MB 152.00 MB 160.00 MB 168.00 MB 176.00 MB 192.00 MB 312.00 MB 480.00 MB 536.00 MB 544.00 MB 568.00 MB 584.00 MB 728.00 MB
nccl_speed(GB/s) 0.21 0.43 0.59 0.81 1.02 1.12 1.31 1.31 1.49 1.67 1.66 1.93 1.90 2.19 2.15 2.50 2.89 2.98 2.76 3.62 3.77 2.60 3.33 4.16 5.24 4.13 4.76 4.32 4.35 5.61

@ZeldaHuang ZeldaHuang merged commit 28d2743 into main Oct 16, 2024
14 checks passed
@ZeldaHuang ZeldaHuang deleted the async_llumlet branch October 16, 2024 13:03
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants