Comware Based
cancel
Showing results for 
Search instead for 
Did you mean: 

5920 switches with IRF stack / using for iSCSI / seeing lots of packet drops

 
Highlighted
Occasional Advisor

5920 switches with IRF stack / using for iSCSI / seeing lots of packet drops

Hi folks

I'm looking for some general best-practise type info regarding using these 5920's for iSCSI.  They are dedicated to iSCSI so no need to factor in mixed workloads.

We have four 24-port switches in two IRF stacks (iSCSI-A and iSCSI-B for each connected host).,  The IRF stack ports are aggregate pairs I think, giving a total of 20GbE in each direction.  Certainly there are four ports on each switch that are participating in the IRF links, and there's only two switches per IRF stack.

I would prefer not to share config at this point, I'm really after generic guidelines rather than specific configurations.

Questions:

  1. Should the IRF stack ports have flow control enabled?  Flow control is disabled everywhere else (in line with storage vendor's recommendation)
  2. We are using the standard/default port buffer allocation strategy.  I see that there is a "burst mode" QoS policy available, and we are definitely dropping some packets when the individual port buffers are full.  The overall buffers don't appear to become exhausted.  Burst Mode appears designed for exactly this scenario - can anyone confirm?  (I'm somewhat concerned about enabling any QoS policy as it comes with a big warning that once enabled it cannot be removed!)
  3. Jumbo frames are enabled and used on every port.  Are there any other (either global or port-specific) configurations that are generally a good idea for iSCSI storage network infrastructure?

Thanks heaps, any assistance really appreciated.

-Jay

1 REPLY 1
Highlighted
HPE Pro

Re: 5920 switches with IRF stack / using for iSCSI / seeing lots of packet drops

Hello Jay

The answers for your questions are :

1- No, the flow control shouldn't be used for IRF ports

2- Yes burst mode is designed for burst traffic and can manage instantly burst traffic to avoid dropping.

3- There is no special command for iSCSI usage of port, just as general recommendations, configure STP as edge port or disable it to avoid port blocking in case of topology change.

Others recommendations also :

- With tow devices use IRF in chain mode to get more bandwidth with the same redundancy level.

- If the ports are used for other traffic than the iSCSI it's highly recommended to use Qos.

Hope these can help you to optimize your infrastructure.

best regards

I am an HPE Employee

Accept or Kudo