Unsolved

This post is more than 5 years old

111 Posts

2281

May 14th, 2009 01:00

Uneven traffic with PowerPath

Hey guys,

I've been looking through our SAN traffic, which contains two separate Fabrics, each with core Cisco MDS9513 (with the DMXs are attached to) and edge MDS9509 switches (with hosts attached to). I've noticed that some windows hosts with PP installed do not split the traffic between the two Fabrics...

I've checked PP, license, policy (SymmOpt), ISLs, HEAT report was ok too. One of the hosts is even a part of a cluster which is connected the same as the other host, but only one of them do not split traffic. Traffic on those "problematic" hosts ranges betweem 5% on 1 fabric and 95% on the other and sometimes 20%-80%... All other 300+ hosts usually split it even at 50%-50%.

Any ideas?

Thanks,
Maxim

6 Operator

 • 

5.7K Posts

May 14th, 2009 01:00

Traffic on those "problematic" hosts ranges betweem 5% on 1 fabric and 95% on the other and sometimes 20%-80%.


Ok, so you ruled out the possibility that perhaps 1 zone is missing or 1 HBA is broken.
Hmmmm

111 Posts

May 14th, 2009 02:00

Yeah. Zones and masking are ok. PP shows both paths alive and active.
What can it be?

2 Intern

 • 

1.3K Posts

May 14th, 2009 03:00

How is the connection of problematic host in the SAN? Does this host hav two core connections or routed thorugh ISL? ISL connections may give load balance problem unless the truncking license is available.

can you post "powermt watch " from the problem host. This will give the IOs over each paths/HBA

111 Posts

May 14th, 2009 04:00

It is connected to the edge switches, while the storage is in the core.
Meaning yes, the traffic goes through the ISLs which are trunked.
But like I said, I have about 300 other servers that are connected the same way.
Some of them are connected to the same FAs.

Somehow, whenever I login to the server, the 'powermt watch' shows 0 i/o. which is weird :)

I'm also troubled that all problematic hosts are windows.
Haven't seen this on Linux, HP, AIX, Solaris or VMWARE... :S

197 Posts

May 18th, 2009 08:00

And you have 4/4 paths for each device? I had an issue where all 4 paths were logged in and registered with the array but only 3 paths were being used on the server. I would see the same utilization as you were.

197 Posts

May 19th, 2009 04:00

Thanks, I completely missed that.

111 Posts

May 19th, 2009 04:00

Not 4/4.
Just 2/2

6 Operator

 • 

5.7K Posts

May 19th, 2009 04:00

He's connected to Symm, not CX, so that would be 2 HBA's, 2 FA's thus 2 zones.

111 Posts

May 19th, 2009 09:00

So does anyone have an idea?

11 Legend

 • 

20.4K Posts

 • 

87.4K Points

May 19th, 2009 10:00

anything funky in HBA configuration ? throttle settings, queue depth, topology ?

111 Posts

May 20th, 2009 00:00

I wish...
Not that simple :)
Both EMCreports and manual check confirmed EMC settings...

6 Operator

 • 

5.7K Posts

May 20th, 2009 01:00

Something weird in the switch ? Perhaps 1 port is set to 1Gb instead of 4 ?

111 Posts

May 21st, 2009 00:00

Actually I forgot to check the speed. Thanks RRR. But this is not the case.
All HBAs sync at 2gbs.

39 Posts

May 21st, 2009 05:00

Are you seeing similar numbers of queued IOs on each path? Or does one of the paths seem to always have a couple of IOs sitting on a path while the other fluctuates?

11 Legend

 • 

20.4K Posts

 • 

87.4K Points

May 21st, 2009 05:00

i know this is not the most desirable way to fix thing but have you tried re-installing PP ?
No Events found!

Top