UCF Hackathon 2019 - openucx/ucx GitHub Wiki
You are registering for an open public standards setting discussion and development meeting of UCF. The discussions that take place during this meeting are intended to be open to the general public and all work product derived therefrom shall be made widely and freely available to the public. All information including exchange of technical information shall take place during open sessions of this meeting and UCF will not sponsor or support any private working group, standards setting or development sessions that may take place during this meeting. Your participation in any non-public interactions or settings during this meeting are outside the scope of UCF's intended open-public meeting format.
December 9-12.
- 9 (Only afternoon session)
- 10-11 (Full days)
- 12 (only morning session)
https://forms.gle/mFTkYgDRtp2hPejy8
ARM, Inc
5707 Southwest Pkwy #100, Austin, TX 78735
Arm parking next to the office building. You can use any un-labeled parking spot.
Arm Austin Lobby - we have two buildings in this area. Lobby/registration will in the building with Arm logo. The actual meeting will be in the second building, first floor (Training room C). Feel free to call/email Pavel Shamis.
Please upload your slides (pdf or pptx) here
| Date | Time | Topic | Speaker/Moderator |
|---|---|---|---|
| 12/09 | 11:30-12:00 | Registration | Pasha/Megan |
| 12:00-13:00 | Lunch | Training C | |
| 13:00-13:50 | InfiniBand and RDMA recent advances | Gilad Shainer | |
| 14:00-15:50 | Proposal for collective operations API and implementation overview | Manju/AlexM | |
| 16:00-17:00 | Hardware specific (u-arch) codes in UCX | AlexM | |
| ------- | ------------- | -------------------------------------------------------------------- | ------------------ |
| 12/10 | 08:45-09:00 | Registration Join over Web Connect or Skype | Pasha/Megan |
| 9:00-9:50 | Breakfast: LANL future directions | Steve Poole | |
| 10:00-11:00 | GPU Affinity | Akshay/Yossi | |
| 11:00-12:00 | GPU Affinity | Akshay/Yossi | |
| 12:00-13:00 | Lunch | Training C. | |
| 13:00-14:00 | Pipelined transfers within the node, am_zcopy for cuda-ipc | Devender | |
| 14:00-14:50 | RDMA-Core State of the Union | Jason | |
| 15:00-15:30 | UCX containers | Mikhail. | |
| 15:30-16:00 | Documentation | Brent/Pasha | |
| 16:00-16:30 | Release Procedure | Yossi/Pasha | |
| 16:30-17:00 | CI at Azure | Jason/Yossi | |
| ------- | ------------- | -------------------------------------------------------------------- | ------------------ |
| 12/11 | 08:45-09:00 | Registration Join over Web Connect or Skype | Pasha/Megan |
| 09:00-9:50 | Breakfast: Charm++ Overview | Nitin | |
| 10:00-10:30 | Charm++ - present work and results | Mikhail | |
| 10:40-12:00 |
|
Yossi | |
| 11:00-12:00 | Lunch | Training C | |
| 12:00-13:00 |
|
Akshay/Devendar/Yossi | |
| 13:00-14:00 | UCP request API | Mikhail | |
| 14:00-14:50 | UCP active message API | Sameh | |
| 15:00-15:50 | Support for different memory types | Yossi | |
| 16:00-17:00 | RapidsAI/Dask/Joins | Nikolay | |
| ------- | ------------- | -------------------------------------------------------------------- | ------------------ |
| 12/12 | 08:45-09:00 | Registration. Join over Web Connect or Skype | Pasha/Megan |
| 09:00-09:50 | Breakfast: SparkUCX and Java API | Yossi | |
| 10:00-11:00 | Smartnic | Pasha | |
| 11:00-12:00 | Open discussion | Pasha | |
| 12:00-13:00 | Lunch and adjourn | Training C |
- EP address bloat - almost 1k today. How we can compress this ? message pack ? Maybe we should switch to on-demand address exchange
- Add CLA bot
- Add requirement for git signatures
- Jason suggest to distribute rpm and Debian packages
- Azure CI FAQ for UCX
- Charm++ Spack installation has to be update
- Charm++ API requests, active messages, priorities
- Follow up with Nathan and Steve on UCT API stabilization / Nathan
- Follow up with board on intermediate progress with UCG proposal / Pasha / Steve
- UCP protocols v2 proposal
- CUDA-IPC cache free
- Profile CUDA-IPC transport
- No nv_peer_mem = crash (probably because IB is detetecting buffer as host)
- #4568
- #4567
- #4566
- #4565
- #4564
Original agenda laundry list
| Topic | Speaker/moderator | Time |
|---|---|---|
| Proposal for collective operations API and implementation overview | AlexM | 2 |
| Hi1620 - hw-specific code in UCX | AlexM | 1 |
| GPU - HCA affinity | Akshay,Yossi | 2 |
| Pipelined transfers within the node, am_zcopy for cuda-ipc | Devendar | 1 |
| Multi-rail for TCP sockets | Akshay | 1 |
| UCX in containers | Mikhail | 0.5 |
| CI and AZP | Yossi,Jason | 0.5 |
| UCP request API | Mikhail | 1 |
| UCP active message API | Sameh | 1 |
| Charm++ - present work and results | Mikhail | 0.5 |
| Support for different memory types | Yossi | 1 |
| RapidsAI/Dask/Joins | Nikolay | 1 |
| RDMA Core | Jason | 1 |
| SmartNIC | Pasha | 1 |
| SparkUCX and Java API | Yossi | 1 |
| Release procedure | Pasha,Yossi | 0.5 |
| Documentation for users | Pasha | 0.5 |
| UCP protocols architecture | Yossi | 1.5 |
| Total | 17.5-18 |