Powerpath Maximum # of paths

Todd;

PowerPath allows for up to 32 paths per logical device. PowerPath exploits the multipathing capability of storage systems. Therefore the 8 HBAs can have 4 paths each to the same device for a total of 32 paths.

Nick

Understanding the 32 paths per device now, if I needed to use more than the same 4 FE’s on the VMAX, so as to better balance across all 8 engines, I would need to break up the provisioning into multiple groups?

Example:

IG=HBA 1,2

PG=Qty 8

SG= ¼ of luns

HBA 2,3

PG=Qty 8

SG=1/4 of luns

Etc for total of 4 groups.

Does that make sense to do in order to maximize 32 paths (2 HBA w/ 4:1 Fanout) ?

- Todd Larson - EMC Residency Practice - 214.394.4860 - Todd.Larson2@emc.com

Exactly, a Host HBA can certainly overrun a VMAX FA so I'm trying to get a 4:1 FA:HBA FanOut.

currently it's 8 FA's to 8 HBA's (1:1)

I want to test 32 FA's to 8 HBA's to spread the load across more Engines / FA which are currently seeing 2000+ IOPS each.

Usually the HBAs can do a lot more IOPs than a VMAX FA, so having more FAs than HBAs isn't an issue with small block IO.  With large block transfers, the pipes can fill up, so having too many FAs to one HBA could make the HBA a bottleneck.

Are experiencing performence degradation on the host? Or do you just want to leverage the max possible 32 paths?

I'm not 100% onboard with instituting 32 paths to one device especially with only 8 HBAs. PowerPath is primarily for path redundancy and load balancing. In my opinion one path from each HBA would be suffieicent.

If you want to test then go for it.  In the meantime have you checked PowerPath to determine which FAs the IOs are travelling?  Are the FAs that the HBAs connect to hitting the 2000+ IOP mark? Are the FAs hosting multiple applications or are they dedicated to one app?  Have you tried Round Robin? I would also suggest runnng an emcGrab on the host and opening a support ticket for an assessment (benchmark) before rezoning.

Currently the host uses Solaris native MPIO and I'm pushing for PowerPath as it has more inteligence in it's optimized load balancing then just plain round-robin.

And yes, it's the dedicated VMAX FA slices which show MAX IOPS >2000

are you expecting an advantage with Round Robin? just curious ..

I'm expecting an advantage OVER round-robin MPIO by using PowerPath.

Should I not be?

Powerpath will schedule more IO to the paths with the best response times and will also distribute workload across all surviving paths should one or more paths fail. So your expectations around improvement over MPIO are good.

PowerPath Round Robin is just a thought.  This non intrusive test would alternate the IOs amongst the FAs and observe the loads.  In any case have you opened a support ticket?  Support will look at the Host logs files, check the VMAX and make recommendations.

dynamox

The discussion is on testing 32 FA's to 8 HBA's to spread the load across more Engines / FA which are currently seeing 2000+ IOPS each.

I was merely making a recommendation for a non-disruptive test before rezoing 32 paths to the FAs.  In addition it is my recommendation to have the host logs and VMAX reviewed as well for possible anomolies or at least a benchmark.  

why would i invest in PowerPath then if native MPIO will do just of a good job doing round-robin. I buy PowerPath for its intelligence.