G A L A X Y ® A U R O U R A C O N F I G U R A T I O N A N D S Y S T E M I N T E G R A T I O N G U I D E
117
Section 4 Troubleshooting Guide
about (6) 16-drive backplanes. If you need more, instead of an LSI 3081e, use
a variant called an LSI 3801e – It looks exactly the same, except instead of
two 4-lane ports connecting to one channel, it has two 4-lane ports, each on a
separate channel.
The internal discreet SATA connectors and especially a sideband connector –
are especially delicate and prone to breakage. The actual card-edge
connection portion of the multilane connector typically isn’t a problem, what is,
is the small metal spring button which secures the cable to the shield of the
connector it is plugged into. This button can and will move or shift. When it’s all
the way back, towards the cable, the position will prevent it from locking into
the shield – it must be all the way forward, and the two latches on it must lock
to the shield in order to be sure that the card-edge connector on the cable is
securely mated properly. If this latch becomes bent, it must be fixed – at all
cost. If it can not be fixed, the cable has to be replaced. If the cable is used
with a broken latch, then it’s possible that not all of the drives connected to the
cable will come up.
4.13
Infiniband HCA problems
These Infiniband HCA cards (Mellanox) are very simple, and very reliable. One
mortal feature is if the cable plugged into them (externally) is pulled too hard, it
might pull the card out of the PCIe slot. However the card itself is mainly
troubleshot through software. There are two LEDs on the card. One for Link,
and the other for Activity. The Link LED comes on when a subnet manager is
running. If it’s off, most likely the subnet manager is not running. If the activity
LED doesn’t blink, then chances are there is no activity. It is better to try to
eliminate the software/cables before pointing to the card. There is a heat-sink
on the card, held in with spring clips. If the spring clips break, the heat sink will
come off and the card may overheat.