Disk path switching

[Steve-Thompson]

I like to hear the forums technical opinion to settle a discussion we have going.

The configuration is a highly redundant cluster using
2 ES40 nodes, each in a different building about 500m apart.

The cluster uses memory channel with extenders to give us the required distance.
We use 7.22 OVMS (Patched) on a HSG (V86S-10 software) SAN.
Each building has it's own ES40, HSG and 2 "8EL" switches.
There are a variety of R1 and R5 disks on each HSG, duplicated between buildings.
The disks are further protected with Disk shadowing to maintain copies of the operating system and data on the disks between buildings.
The configuration results in 4 possible paths to each disk, seen from VMS.

Our discussion is that of using "preferred paths" on the HSG controllers to seperate the disks between paths under VMS....

My viewpoint is.
1.
Preferred paths should not be configured!
(There are prefered paths on the HSG's and despite this VMS sees all disks on the same path after a reboot. Obviously it would appear preferring a path "doesnt do anything").

2.
Paths should be set from the operating system (eg. after every reboot) as required, I/O being the main consideration.

3.
What does the forum say or add to this scenario?

Thanks for your efforts

[Wim Van den Wyngaert]

We distribute the active paths randomly over the available paths (in fact a 1-4 counter that is assigned to the disks). We do this during each boot.

Wim

[Wim Van den Wyngaert]

This is our procedure. Some site logicals may be needed.

Wim

[Steve-Thompson]

THanks Wim

Can you confirm your configuration is at the HSG level, ie preferred_path=
or

Regards

[Steve-Thompson]

... the problem is the information we have.

The manual "Guidelines for OpenVMS CLuster Configurations" (The book is the 7.2-2 version.)

- states (sorry recommends)that "preferred_paths" should not be set.

It supports this by stating that VMS overrides path "preferring" at the HSG level.

HP tech support "recommend" preferring the paths.

Is this really important, I wonder?

We're tuning the system "to get a bit more out of it" .... that's when this all started.
:-)

Regards
Steven

[Jan van den Ende]

Steven,

all drives "behind" an HSG (EMA, EVA) are "local" to VMS.
But some "locals" are more local than others.
Not sure about 7.2-2, but certainly 7.3-2 has the option toe define SITE values.
And then each member can be assigned SITE_2, SITE_2, SITE_3...
SITEs can be assigned different cost, and READ access takes the cost difference in account when choosing the prefered member.
But... the speed of light, even through glass, is such that distances well below 5 KM are negligeable (measured in time) compared to the time spend by mechanical and electronic delays.

Do the math yourself, lightspeed in glass is pretty close to 200 000 KM/sec.

At your distances, it really is NOT worth the effort.

hth

Proost.

Have one on me.

jpe

[Steve-Thompson]

Hi Jan
thanks for your input .... I see where you're going .... but the set device/site isnt available under 7.2-2!
You've touched on as aspect of the "correct" configuration I want to achieve.

However the main "hair-split" is the usage of "preferring" disks at the HSG level.
The book says NO and an official source says "it's better to" ...
(It's like going to a second doctor for his opinion)

I "prefer" the book's guidelines which coincides with obervation. VMS certainly ignores the "preferring" of disks as ALL disks start up on the SAME path regardless of the HSG configuration.

Steven

[Jan van den Ende]

Steven,
I would have expected "HP tech support" to recommend upgrading!

Fiber disks were first introduced at 7.2-2 (IIRC that version was explicitly released to introduce Fiber disk!!).
And there were still quite some things that they did not yet have, or have not entirely right, a situation that has much improved with the various upgrades and patches.

again, fwiw

Proost.

Have one on me.

jpe

[Steve-Thompson]

Hi Jan

I agree with upgrading however, it'll be a process of about 6 months to do so!

Thanks for playing ball!

Steven

[Wim Van den Wyngaert · Accepted solution]

We do not have prefered paths on HSG level.
And we use SITE. But SITE only concern on which disks it will read (the ones in the building of the node). For write there is of course no improvement.

Wim

[Robert Brooks_1]

Beginning with VMS V7.3-1, multipath attempts to spread the "current" path on all disks across all available local paths during device configuration (usually at boot time). V7.3-1 is also the first release to support the use of the remote (MSCP) path. If you are to upgrade, I recommend (in decreasing order) V8.3, V8.2, or V7.3-2. The most recent versions have been kept current with respect to multipath and the
SCSI and fibre channel device drivers.

The SITE construct is only useful for multisite clusters with shadowing; it directs host-based volume shadowing to do reads from the local disk, as opposed to the default round-robin algorithm that alternates reads across all members of a shadow set. The SITE construct has nothing to do whatsoever with multipath local paths.

-- Rob