[FEA] NCCL integration

[BradReesWork]

We are probably going to need to integrate NCCL into Hornet to support multi-gpu analytics.

• Update CUDA Code
• Code document
• Test cases

[ogreen]

We need think about general types of communication patterns used in multi-processor graph algorithms and find a way to wrap them up so that NCCL can be used by non experts.

[ogreen]

This task will play a key-role in our ability to design multi-gpu traversals.

[seunghwak]

OK, and I guess this is related to what we (you, Aatish, and me) have discussed in the previous meeting (Building a framework/API to accelerate common patterns in GPU graph algorithms).

…

-seunghwa From: ogreen <[email protected]> Reply-To: gpuopenanalytics/graph-research <[email protected]> Date: Thursday, January 3, 2019 at 11:44 AM To: gpuopenanalytics/graph-research <[email protected]> Cc: Seunghwa Kang <[email protected]>, Assign <[email protected]> Subject: Re: [gpuopenanalytics/graph-research] [FEA] NCCL integration (#13) This task will play a key-role in our ability to design multi-gpu traversals. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub<#13 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/Aru6kQHF51aDYU8wSaMyJZ2djE3naaClks5u_l2MgaJpZM4Zof0g>.

----------------------------------------------------------------------------------- This email message is for the sole use of the intended recipient(s) and may contain confidential information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message. -----------------------------------------------------------------------------------