Mellanox, the vendor for the InfiniBand cards in each Pleiades node, has developed a firmware fix for the InfiniBand-related queue pair (QP) errors that caused some jobs to fail over the past several months. Testing at NAS and SGI shows that the new code eliminates the QP error.
Starting at 5:00 p.m. Thursday, September 4 (Pacific Time), we began rolling out the firmware to the nodes. Jobs starting after that time will use the new firmware.
For more details and important recommendations, please see the following Knowledge Base article:
If you have questions or concerns, please contact the NAS Control Room at (800) 331-8737, (650) 604-4444, or support@nas.nasa.gov.