VA7410 - Most efficient path

[Kurt Henning]

We just installed a VA7410, and I'm trying to figure out how to configure the VGs for Oracle. The conventional wisdom on VA7410s seems to be that the RG1 controller (lefthand side when looking at the back of the VA) is the most efficient path to RG1, while the RG2 controller (righthand side) is the most efficient path to disks in RG2.

That makes sense on paper, but makes for confusion when I look at the back of the VA. Each of the controllers on the VA has two inputs (H1, and H2). We have two interfaces installed in each of our production servers. Each interface goes to one of two switches. The reason for the confusion is this:

The first switch has connections to both H1 on the RG1 interface, and H1 on the RG2 interface. The second switch is similarly configured with connections to H2 on RG1, and H2 on RG2. Here's a partial ioscan that might help in visualizing this.

Class I H/W Path Driver S/W State H/W Type Description
====================================================================================
fc 2 0/4/0/0 td CLAIMED INTERFACE HP Tachyon XL2 Fibre Channel Mass Storage Adapter
/dev/td2
disk 96 0/4/0/0.1.0.0.0.0.2 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c14t0d2 /dev/rdsk/c14t0d2
disk 103 0/4/0/0.1.0.0.0.0.3 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c14t0d3 /dev/rdsk/c14t0d3

disk 98 0/4/0/0.1.1.0.0.0.2 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c16t0d2 /dev/rdsk/c16t0d2
disk 104 0/4/0/0.1.1.0.0.0.3 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c16t0d3 /dev/rdsk/c16t0d3


fc 3 1/10/0/0 td CLAIMED INTERFACE HP Tachyon XL2 Fibre Channel Mass Storage Adapter
/dev/td3
disk 100 1/10/0/0.1.0.0.0.0.2 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c18t0d2 /dev/rdsk/c18t0d2
disk 105 1/10/0/0.1.0.0.0.0.3 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c18t0d3 /dev/rdsk/c18t0d3

disk 102 1/10/0/0.1.1.0.0.0.2 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c20t0d2 /dev/rdsk/c20t0d2
disk 106 1/10/0/0.1.1.0.0.0.3 sdisk CLAIMED DEVICE HP A6218A
/dev/dsk/c20t0d3 /dev/rdsk/c20t0d3

It depicts two VA LUNs. There are a total of four possible device paths for each LUN, so which one is most efficient? In the first pair c14t0d2 is on RG1, and c14t0d3 is on RG2. So:

c14t0d2
c16t0d2
c18t0d2
c20t0d2

All four of the above device files are for the same LUN in RG1, so which represents the most efficient path?


c14t0d3
c16t0d3
c18t0d3
c20t0d3

All four of the above device files are for the same LUN in RG2, so which represents the most efficient path?

If the 0/4/0/0 interface had a more direct path to RG1, and the 1/10/0/0 interface had a more direct path to RG2 it would be easier to figure out. But, the way the cabling is configured, each interface has equal access to the RG1 and RG2 interfaces because the 0/4 interface connects (through the switch) to H1 on RG1 and H1 on RG2. The 1/10 interface (through the other switch) connects to H2 in RG1 and H2 on RG2.

Does this mean that it doesn't really matter which device file is designated as the primary, and which ones are used as "alternate links"?


-Kurt Henning

[Eugeny Brychkov · Accepted solution]

Kurt,
you'd better post whole ioscan or at least 'ioscan -fnCdisk' output.
1. where's LUN 0? I do not see it in the list. Without LUN 0 you will not be able to manage VA using CommandView software;
2. let's look at this path
0/4/0/0.1.1.0.0.0.2
0/4/0/0 is hardware path where FC HBA is installed;
1 - switch domain id (both have the same '1', so it's hard to determine who is who);
1 - switch port number. For 3x00 it's port #1;
0.0.2 - LUN number, in this case LUN 2.
3. host ports, H1 and H2, are 'equal rights' ports. Primary path should go to C1 for RG1 and C2 to RG2 through any of these ports. So to choose right primary path you should do the following:
- select LUN (let's say, LUN 1);
- check in which RG is it (let's say, RG1);
- look at C1 and choose one of VA frontend ports, then note to which switch and which port of this switch this VA port is connected (let's say, port 1 switch domain #1).
Thus now you know that part of HW address looks like
1.1.0.0.1
- you go to this switch and check to which host HBA this switch is connected (let's say, 0/10/0/0).
Thus right hardware path is
0/10/0/0.1.1.0.0.1
- you look at ioscan output and match this hardware path to device file and then assign this device file as a primary path for LUN (hpux PV)
Eugeny

[Roger_22]

LUN 0 is not required for hpux - it makes life easier, but not required - Command View will find and use any LUN that it has acces to send commands.

Managing the ???performance path??? is less of an advantage with the new VA7410 and 7110 then the earlier products ??? and for both generations, the gain will only be modest for typical workloads.

Managing the performance path is good, but it???s not the magic key to performance. (There is more performance value in assuring the load is balanced between the two RGs)

An IO request thru the non-performance path will add an additional ~<30us (microseconds) to the response time. While this is measurable for cache hits, it is insignificant for IOs that will access a disk. Remember, the average access time for a disk is about 8,000 to 10,000us (8-10milliseconds)

But, if you???re eccentric then here???s one for you: For the 7410/7110, the IPC (inter-processor call ??? what happens when you send the command request to the ???non-primary controller???) takes fewer microprocessor cycles than managing the fibre channel hardware. So, if you are using just one RG (always in the 7110, and possible in the 7410), and at full utilization (no idle microprocessor cycles), sending the commands to the non-primary controller will yield the maximum IOPS performance!

Given all this, here my recommendations:

If you stripe the access to the array so that both RGs get the exact same workload, then there may be a slight advantage to using the performance path. If both RGs are not guaranteed to get the exact same workload, then it doesn???t matter. In any case, the gain will only be small.

[Kurt Henning]

Thanks for the advice so far. I didn't include all of the ioscan, because it's not really relevant. There is indeed a lun 0, but its hardware path is almost the same (it just ends in 0 )as for the "disks" I did include.

Since we have two HBAs in each server, and each HBA has a path to one of the host interfaces on each of the RG controllers, I was just stumped about how to determine the "performance path", or even if there WAS a performance path in this configuration.

I know the usual advice is to pick the device file that corresponds to the HBA that has the most direct path to the VA controller governing the RG in question, but in our case each HBA has equal access to both RG controllers, so I'm not sure a "performance path" even exists. I am, however, new to the VA world so I wanted to make sure I wasn't missing something.

I just don't want to pick a poorly thought out configuration, then have to tear down the VGs and start over at some future date in an effort to enhance performance.