[openib-general] How do we prevent starvation say between TCP over IPOIB / and SRP traffic ?

[Richard Frank]

Currently, all I have is problem to resolve. We will think about a
general model..

For Oracle to support running RAC on a single fabric - assuming the
fabric is utilized for both network (inter node cluster com) plus
storage I/O - we need to limit /control latencies of cluster network
msgs such that increasing the storage I/O load does not impact the QoS
requirements for cluster network comm.

A globally set service level / QoS for cluster network traffic such that
it has better / more stringent QoS requirements than storage I/O traffic
would meet our needs. 

However, having the capability to define service levels on a per
connection or ULP or process basis is interesting too. 


On Wed, 2006-04-19 at 09:44 -0700, Caitlin Bestler wrote:
> openib-general-bounces at openib.org wrote:
> > Some application level protocols - require higher QoS levels than
> > others - for various communication and I/O operations.
> > 
> > For example, cluster inter-node health msgs have fixed
> > latency requirements that if exceeded may result in
> > unexpected node removals from the cluster.
> > 
> > Are there any mechanisms available to the client process to
> > manage the QoS level for the various supported ULPs
> > (SDP,TCP,UDP,RDS,SRP,iSER,etc) either at the ULP level or
> > some combination of process and ULP - or perhaps even at the
> > connection level ? 
> > 
> > Using the same example, the cluster node monitors might set
> > the priority / QoS level of the heart beats to be more
> > important than normal SRP/iSER traffic to ensure no starvation ?
> > 
> > 
> 
> Working up from hardware capabilities and trying to generalize
> them probably won't lead anywhere.
> 
> Do you have a model of the requirements for transport/device
> neutral QP prioritization that would meet your needs?
>