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.

Sign up for GitHub

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

Jump to bottom

Client/Server use wakeup feature and their own thread #27

Open
wants to merge 36 commits into
base: master
Choose a base branch
from
Open

Client/Server use wakeup feature and their own thread #27

wants to merge 36 commits into from

Conversation

jeynmann
Copy link

@jeynmann jeynmann commented Jul 17, 2023 •
edited
Loading

  1. Use wakeup feature for clients/servers' worker.
  2. Clients/servers use their own thread to progress and send/reply message.

The following table is the testing results on the customer's cluster.
(Patch means using the spark-ucx of this branch.)

mode sparkucx_version shuffle data shuffle read stage
DC_X/TCPCM master 559.4 GB 0.15 min
DC_X/TCPCM master 559.4 GB 0.12 min
DC_X/TCPCM patch 559.4 GB 0.08 min
DC_X/TCPCM patch 559.4 GB 0.08 min
TCP 559.4 GB 0.08 min
TCP 559.4 GB 0.08 min
DC_X/TCPCM master 1118.8 GB 0.2 min
DC_X/TCPCM master 1118.8 GB 0.18 min
DC_X/TCPCM patch 1118.8 GB 0.15 min
DC_X/TCPCM patch 1118.8 GB 0.13 min
TCP 1118.8 GB 0.2 min
TCP 1118.8 GB 0.2 min
DC_X/TCPCM master 2.2 TB 0.3 min
DC_X/TCPCM master 2.2 TB 0.33 min
DC_X/TCPCM patch 2.2 TB 0.3 min
DC_X/TCPCM patch 2.2 TB 0.25 mi
TCP 2.2 TB 0.48 min
TCP 2.2 TB 0.3 min
DC_X/TCPCM master 3.3 TB 0.53 min
DC_X/TCPCM master 3.3 TB 0.53 min
DC_X/TCPCM patch 3.3 TB 0.38 min
DC_X/TCPCM patch 3.3 TB 0.43 min
TCP 3.3 TB 0.58 min
TCP 3.3 TB 0.6 min
DC_X/TCPCM master 4.4 TB 1.1 min
DC_X/TCPCM master 4.4 TB 1.1 min
DC_X/TCPCM patch 4.4 TB 0.73 min
DC_X/TCPCM patch 4.4 TB 0.7 min
TCP 4.4 TB 0.77 min
TCP 4.4 TB 0.75 min
DC_X/TCPCM master 5.5 TB 1.4 min
DC_X/TCPCM master 5.5 TB 1.4 min
DC_X/TCPCM patch 5.5 TB 0.93 min
DC_X/TCPCM patch 5.5 TB 0.95 min
TCP 5.5 TB 0.98 min
TCP 5.5 TB 1.0 min
DC_X/TCPCM master 10.9 TB 2.4 min
DC_X/TCPCM master 10.9 TB 2.4 min
DC_X/TCPCM patch 10.9 TB 1.6 min
DC_X/TCPCM patch 10.9 TB 1.6 min
TCP 10.9 TB 1.8 min
TCP 10.9 TB 1.9 min

JeynmannZ and others added 30 commits April 25, 2023 07:13
@JeynmannZ
bind worker to thread
084fd6d
@JeynmannZ
use wakeup/yield in server/client worker
761792d
@JeynmannZ
use asyn
87b19e0
@JeynmannZ
use loop
c075505
@jeynmann
Merge branch 'NVIDIA:master' into bind_worker_wake_asyn
760a276
@JeynmannZ
connect executors in setup thread
fb0bb34
@JeynmannZ
Merge branch 'con_in_setup' into bind_worker_wake_asyn_conn
2f44bc0
@JeynmannZ
yield in loop
6b4d43e
@JeynmannZ
reverse client bind
d6c1978
@JeynmannZ
replace while with if in client progress
731d5d8
@JeynmannZ
Merge branch 'bind_worker_wake_asyn_conl' into bind_worker_wake_asyn
8e49765
@JeynmannZ
remove useless modifications
1279bc8
@JeynmannZ
use sync send
a1e09ec
@JeynmannZ
bind worker to thread
9267414
@JeynmannZ
bind client worker
f94359f
@JeynmannZ
use rr instead of blocked queue to select client
4c3ba6e
@JeynmannZ
add progressBlocked for flush
d8c12f1
@JeynmannZ
add asyn
893ee5c
@JeynmannZ
add pre connection
1546d73
@JeynmannZ
submit req after fetch
3f14725
@JeynmannZ
bind client worker to fixed thread
931fba1
@JeynmannZ
temp
6a49fd6
@JeynmannZ
split context
e393aa3
@JeynmannZ
client use monitor
7c47170
@JeynmannZ
rm monitor
cdc9992
@JeynmannZ
add sync in worker thread
b04c225
@JeynmannZ
connection monitor
8267cf7
@JeynmannZ
reduce synchronized region
1b5ffd5
@JeynmannZ
rm use code
4b95547
@JeynmannZ
reset RR code
a608bb1
@jeynmann jeynmann mentioned this pull request Jul 17, 2023
Open
@jeynmann jeynmann changed the title Client/Server use their own thread [DNM] Client/Server use their own thread Jul 17, 2023
@jeynmann jeynmann changed the title [DNM] Client/Server use their own thread [DNM] Client/Server use their own thread and wakeup feature Jul 17, 2023
@jeynmann jeynmann changed the title [DNM] Client/Server use their own thread and wakeup feature [DNM] Client/Server wakeup feature and use their own thread Jul 17, 2023
@jeynmann jeynmann changed the title [DNM] Client/Server wakeup feature and use their own thread [DNM] Client/Server use wakeup feature and their own thread Jul 17, 2023
@jeynmann jeynmann mentioned this pull request Aug 8, 2023
Open
@jeynmann jeynmann changed the title [DNM] Client/Server use wakeup feature and their own thread Client/Server use wakeup feature and their own thread May 30, 2024
@jeynmann jeynmann force-pushed the client_server_use_own_thread branch from 53ce91c to 2d1c464 Compare June 3, 2024 08:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@jeynmann @JeynmannZ