[openib-general] [PATCHv2][RFC] kDAPL: use cm timers insteadof own
James Lentini
jlentini at netapp.com
Thu Jun 2 07:36:50 PDT 2005
On Tue, 31 May 2005, Tom Duffy wrote:
> On Tue, 2005-05-31 at 14:34 -0700, Sean Hefty wrote:
>>> Sean,
>>>
>>> Is there any way of requesting an infinite number of retries?
>>
>> There is not, but nothing prevents a user from simply re-issuing a request
>> after it times out.
>
> Infinite retries inside the kernel does not sound like a good idea. How
> would you break it? At least we should have some sort of exponential
> backoff to prevent flooding the network.
We want a value that the consumer can pass as the timeout that will
give the protocol enough time to connect regardless of network
topology. If the value is a fixed period of time, say 500 us, there
will be some network configuration that it won't work on.
The infinite value allows a user to give the CM protocol all the time
it needs to work. If it fails for some other reason (like the remote
node is down) then the connection attempt should fail.
So I think we should do a few things:
- rename this constant
- define its value as 0
- upon seeing it, assign a CM timeout value that allows for a the
protocol to complete
We don't need to do all that on this first pass though. Let's just get
the timeout value working in the general case and then we can worry
about DAT_TIMEOUT_INFINITE and what it means.
james
More information about the general
mailing list