[ofa-general] IPOIB/CM increase retry counts

[Pradeep Satyanarayana]

Jack Morgenstein wrote:
> On Thursday 14 February 2008 09:14, Pradeep Satyanarayana wrote:
>>> On the other hand, if the remote host is actually down, you will make "retry storms"
>>> worse by retrying both at the link layer AND at the TCP layer (each TCP retry resulting
>>> in multiple lower-layer retries).  This will have an effect on the fabric.
>> If the remote host is down establishment of an RC connection does not arise. The UD
>> connection itself will fail.
>>
> I'm talking about the case where the remote host fails after the connection is established.
> In this case, RC will continue to retransmit until the rnr_retry count is exhausted.

Remote host failing should itself be a rare occurrence, and the value for retry=3 is small
enough that it should have minimal effects if any. More over that should have only a transient
effect.


Pradeep