[LU-5798] lnet: router buffering improvements Created: 23/Oct/14  Updated: 13/Feb/19  Resolved: 13/Feb/19

Status: Resolved
Project: Lustre
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Improvement Priority: Minor
Reporter: Isaac Huang (Inactive) Assignee: Isaac Huang (Inactive)
Resolution: Low Priority Votes: 0
Labels: None

Rank (Obsolete): 16263

 Description   

There's a couple of ways to improve router buffering, especially under network congestion:

  • When router is about to forward a message, it checks peer health information, and drops the message if the destination appears dead. In addition, we should also check the source of the message: if the source appears dead, the message should be dropped as well. This should be easy to implement.
  • When under network congestion, a message can get queued indefinitely. It doesn't make sense to continue to buffer the message once the client/server has already aborted the RPC/bulk transfer. The tricky part is to figure out when to drop a queued message. This needs some investigation.


 Comments   
Comment by Andreas Dilger [ 13/Feb/19 ]

Likely already fixed by LNet Health.

Generated at Sat Feb 10 01:54:37 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.