02-03-2011 05:04 PM
Let's say the scenario is a blade chassis with four (4) 10G CEE uplinks to a ToR FCF....
In such a converged network environment at the server edge, how does one determine how much uplink bandwidth to dedicate to FCoE traffic? Is there a standard recommendation to dedicate bandwidth using some sort of policing or rate limiting mechanism or ....what?
I have read a lot about FCoE lately, but I have never seen anything about this.
Any help would be greatly appreciated.
02-04-2011 10:34 AM
There is a mechanism to provide a minimum guaranteed bandwidth for traffic types via Enhanced Transmission Selection (ETS) in the 802.1Qaz standard. This is included in the DCB extensions. This provides a means to set a minimum bandwidth guarantee. I said minimum because if a traffic burst should need more bandwidth than the minimum and IF there is bandwidth available, then it is used by the traffic class. However, when there is no unallocated bandwidth available, then the traffic flow is limited to the "guarantee" level for that traffic class.
That's nice isn't it?
But you ask the hard question, "... how does one determine how much uplink bandwidth to dedicate to FCoE..". This "depends" as it is defined by the cummulative demand for IO to storage of all the applications whose traffic is flowing on that link.
Typically, a designer gets some data from the test environment many companies use prior to deploying an application, or by SWAG, to establish the expected storage traffic rate. The good news with ETS is that you can change the setting for a traffic class if needed. For example, you may decide to reduce the bandwidth for the IP traffic and provide more for storage based on monitoring performance of the application when its in production and seeing that you really out to add another 500 MB/s for the storage minimum bandwidth guarantee. Now, as I said earlier, if there is a burst or spike in a class of traffic, and there is unallocated bandwidth on the link, then the storage traffic can use it when there's a spike. That's very helpful to accomodate the bursty nature of storage traffic.
Mel, I hope this helps.
02-04-2011 10:46 AM
Brook, I know I ask the tough questions. The reason is that I'm a smart guy who can usually answer my own questions. So, its only the impossible questions that I bug you and Chip to answer! LOL
But seriously, I know this is a tough question because one really has to understand the storage demands of each application, or the cumulative demand in the data center, etc...its similar to the question of how much BW to use for client-server traffic and what kind of oversubscription ratios to architecxt for....its tough....
I thought that maybe there was a sort of rule of thumb with regard to expectations in a converged fanric environment with regard to how much of the BW will be used for storage.....