Ticket #27 (new enhancement)

Opened 14 years ago

Last modified 14 years ago

Failed CN can hang the whole pset

Reported by: iskra Owned by: iskra
Priority: minor Milestone: V1R4 Release
Component: ZeptoOS Version:
Keywords: Cc:

Description

If a compute node hangs for whatever reason, and we keep sending it packets on the tree from the ION, this will quickly lock the tree network. That's because packets will back up to the send FIFO on the ION, blocking any other packets to remaining, operational compute nodes.

It is apparently possible to reconfigure the tree interface into loopback mode to read those blocking packets and so unlock the tree on ION. We should ask IBM for the details and implement this.

Change History

comment:1 Changed 14 years ago by anonymous

  • Type changed from defect to enhancement

comment:2 Changed 14 years ago by iskra

  • Owner changed from zepto team to iskra

comment:3 Changed 14 years ago by kazutomo

  • Milestone set to V1R4 Release
Note: See TracTickets for help on using tickets.