Monitoring the InfiniBand Switch

eG Enterprise offers a dedicated InfiniBand Switch monitoring model which periodically checks the operational status of ports and fabric elements and traffic handled by the subnet manager, so that unavailability and abnormal traffic flow of packets can be detected before any irreparable damage occurs.

Figure 1 : The layer model of the InfiniBand Switch

Every layer of Figure 1 is mapped to a variety of tests which connect to the SNMP MIB of the target InfiniBand Switch to collect critical statistics pertaining to its performance. The metrics reported by these tests enable administrators to answer the following questions:

  • What is the current operational status of each fabric element?
  • Is Element manger available for each fabric element?
  • How well the data and unicast/multicast packets were transmitted/received through each PMA Extended port?
  • Which PMA port is symbol error-prone?
  • Which PMA port discarded maximum number of inbound/outbound packets?
  • How many packets with errors were received on each PMA port ?
  • What is the current status of SMA link on a port?
  • What is the current speed of active SMA link?
  • What is the maximum transmission unit (MTU) enabled for each SMA port?
  • Howl well the subnet manager is handling traffic of SMP, GMP and SMInfo packets?