Hi

We are performing some stress tests as preparation of a Oracle DWH environment.

The whole configuration is:

VMAX 20K ( is an older model assimilated to 20K ). 5 engines. FC 600 10K disks, SATA 2TB 7200, and EFD 200gb.FAST is configured.

The DWH volume configuration ( TDEV ) for testing is  12 500GB striped metas bound to FC pool. NO FAST policies applied to them.

From the FA point of view, we have 8 ports dedicated, and connected to 2 CISCO MDS 9513 ( one per fabric ), that own a linecard X-DS9248-96K9, that support 8GB. The port speed has been configured as fixed 8G

From the host point of view, we have a cluster of 2 members that are connected too to the CISCO MDS 9513, and the same linecard. Every host in every fabric have 2 initiators. PowerPath with the SymmOpt policy is configured there.

Port configuration in the linecard is as follows:

1,7,13,19 : HOST

25,31,37,43: VMAX

It means that only the first port of every port group is used. All ports are configured to 8GB and dedicated, so, as per CISCO documentation, we have warranty about that all ports have reserved the 8GB. In fact, in nominal values, we should obtain more or less 800-850MB/s in all ports

Based on this configuration, should obtain a similar rate when use the FIO Linux tool,, but, we obtain a surprising low bandwith on the Symm side:

FIO CONFIGURATION:

[global]
bs=1024k
direct=1
rw=read
ioengine=libaio
iodepth=64
zonesize=4g
zoneskip=4g

[/dev/emcpowerkn]
[/dev/emcpowerko]
[/dev/emcpowerkp]
[/dev/emcpowerkq]
[/dev/emcpowerkr]
[/dev/emcpowerks]
[/dev/emcpowerkt]
[/dev/emcpowerku]
[/dev/emcpowerkv]
[/dev/emcpowerkw]
[/dev/emcpowerkx]
[/dev/emcpowerky]

1 host - 1 storage : 350-375MB /s. This is similar result that we obtain when used 4GB ports.

1 host - 2 storage : 700MB/s. It means that the host side is able to go near to the nominal bandwith, and in this case, powerpath is dividing the storage bandwith ( 350+350 )

1 host -3 storage: near to 800MB/s. 270MB/s obtained in the symmetrix ports

1 host -4 storage: near to 800MB/s. 200MB/s obtained in the symmetrix ports

Observed that FA port %busy is low, FA % busy is low ( below 30% ), but response time is growing up to 300 ms.

When repeat the same test using 2 host ports, the limit in the symmetrix ports is always in 375-400MB/s, never go up.

Confirmed with CISCO team that there is no restriction on por side.

In fact, before to use the 8 FA dedicated FA ports in the 8G linecard, performed same test using 16 shared FA ports connected to 4G switch ports. Configuration was 1 host -- 2 FA, and obtained the desired result: 700MB/s per port, 5400-5600MB/s global, so we are not observing backend limitation.

Opened SR's with CISCO and VMAX teams, but didn't found the required answers..

So, the questions.....

What's the nominal bandwith that should expect from VMAX side? Why we are not getting it?

All collaboration will be appreciated