Are Perf problems in SAP related to switch architecture ?

Unanswered Question
Apr 14th, 2008
User Badges:

Here is an interesting problem on hand that I need some insight into. We had a 2Gb MainFrame 2064 Ficon channel reading data from a 4Gb disk. The mapping was 1:1. The data was all SAP. There was a IBM 2109-M48 Ficon director in between with its port at auto-auto and default Buffer2Buffer credits set to 8. We experienced a huge performance dip. IBM SE's increased the B2B credits to 26 on all ports connected to the channel and disk and the problem went away. At first glance this clearly looks like a problem with the switch buffers. But there are some details that need to be looked into before one can say for sure the switch is a problem. Was the channel unable to process the data on time and hence could not release the buffers back to teh switch causing a backlog ? Was this anything to do with the type of data (SAP) that was being read ? Could getting the disk to speak at 2Gb have helped the situation ? What really happens when once increases the B2B credits dynamically ? Could this action have cleared any congestion the switch was experiencing ? Would I have seen this problem if I had a MDS instead ? If not, why ?

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
inch Fri, 04/18/2008 - 14:21
User Badges:
  • Bronze, 100 points or more

G'day,


Was the disk array performing any replication?


Cheers


stephen2615 Sat, 04/26/2008 - 21:47
User Badges:
  • Bronze, 100 points or more

I have seen numerous problems with Brocade switches left in auto mode. Always set all ports to a desired speed no matter what. I am talking about serious issues where link resets will cause the fabric to keep trying to reconfigure its paths. Another thing to watch for is having the port still online with nothing at the other end actually working but still there.


I am not a fan of Brocades buffer credit handling at all.


Stephen


usabnis Tue, 04/29/2008 - 12:37
User Badges:

Thanks for the insight. The more I study the Problem the more it looks like the switch's internal architecture is not robust to handle load balancing of traffic across its various internal links. Any congestion on the host or target ends up reflecting on the network.

stephen2615 Wed, 04/30/2008 - 14:13
User Badges:
  • Bronze, 100 points or more

There are settings that you can apply that may stop or reduce some of the problems you see.


If you are using HDS storage, contact them as they have some very specific settings that should be applied when you use Brocade switches. As a matter of fact, even IBM say the same thing. An example setting is related to aptPolicy.


Stephen

usabnis Thu, 05/01/2008 - 09:11
User Badges:

The storage we had at the time of the problem was EMC. However it is worth knowing that there are settings to be looked into depending on the type of storage we have.

I

Any idea how the Cisco MDS works in the Ficon environment ?

stephen2615 Thu, 05/01/2008 - 23:08
User Badges:
  • Bronze, 100 points or more

I don't think it makes any difference what storage you use. I think that Brocade needs to be a lot more open with their settings for very important things. I have an account on their website and it is downright painful to get any information out of it compared to the Cisco site which is a breeze. MDS and Ficon works like a charm.

Actions

This Discussion

 

 

Trending Topics: Storage Networking