Compatibility of 8GB FC HBA (in I2 blade) with 4GB FC VC Module.

[The Brit]

Observing an FC IO performance issue after upgrade to 8.4, and BL860c I2.

Old setup:    BL860c,   OpenVMS 8.3-1H1,  4GB dual-Port HBA connected to 4GB FC VC Modules (FC Port 1 to Module 5, and FC Port 2 to Module 6) in c7000 enclosure.

new setup:   BL860c I2,  OpenVMS 8.4,  8GB dual-port HBA, (Enclosure and FC VC modules remain the same.

Upgraded OpenVMS 8.3-1h1 to OpenVMS 8.4 - Ran for several Weeks with No Problems.

Hardware upgrade:

Shutdown system

Deassigned the VC Profile

Swapped out the Blades.

Reassigned the Profile.

Booted new blade.

All OK.

Problem:    I am doing the certification of the new OS/Hardware on my Test System.    The system is set up to mimic Production, i.e.. Same system parameters, same User parameters, pretty much same everything.    However.....

I have a batch job which is running on all of my systems, Production, Development, and now my Test system.      On Production and on my Development systems, the job takes ~20 minutes, however on my new Test system it takes 2.5 hours.    Comparing the accounting information at the end of the jobs shows everything to be similar, except "elapsed time"

I should mention that the new blade has 50% more memory than any of the other systems (400% more than development).     The Development systems are in the same enclosure as the Test system, and are provisioned from the same EVA as the Test System.

The only things which a clearly not the same are.     OS is 8.4 (compared to 8.3-1H1),  Hardware is 860c I2 (compared to basic 860c), and the I2 blade has 8GB HBA's talking to 4GB modules (compared to the old 4GB HBA's).

Is anyone aware of any issues with running mismatched FC Speeds (note FC VC Modules are set to "AUTO").   Or can anyone suggest any other line of investigation that I might check out.     I am running out of ideas.

In the interest of "full disclosure", I have to admit that I can't swear that this wasn't the situation before either OS or Hardware upgrade, except to say that if the performance was this bad, I think I would have noticed it.    After the upgrade it only took two days for the effects to become pretty obvious, unfortunately my logs don't go back far enough, (I purge after 3 weeks.)

I could narrow it down a bit by rolling back both the OS and Hardware upgrades, but I would rather not unless I have to.

I'm hoping one of you might hit on the magic bullet that will fix the problem.

Thanks for any suggestions.

Dave.

[Volker Halle]

Dave,

even if you don't have the old .LOG files, you may still have the old accounting data to compare with.

There are a couple of things, you could try with the FC$SDA SDA extension: see ANALYZE/SYS and FC for some help.

Assuming disk IOs via the FC are much slower now, you could try switching the FC pathes.

SDA> FC PERF can provide very detailled FC disk performance data. 

Volker.

[Colin Butcher]

Hi Dave,

Purely out of interest - what did you discover and what (if anything) improved the situation?

Cheers, Colin.

[EdgarZamora]

I'm interested on an update on this too.  Thanks.

[The Brit]

Hi guys,

        My apologies for not updating at the time, however the problem had nothing to do with the HBA's in the end.     Now here's my second apology, I really don't remember what the details of the solution were.

All I can say is that the cause was something quite mundane.    I "think" it was related to the amount of pagespace that was available on the test system.      I discovered that the same problem occurred in my development systems (still on 8.3-1h1), when I ran the same batch jobs (which didn't normally run there).       Secondly, the problem appears to have been occurring in my test system, prior to the upgrade, without my noticing.

The only place that didnt experience the problem was Production, however production has an obscene amount of page space available (multiple large pagefiles).

Sorry I couldn't reveal something more exciting.     (we decided in the end to forego the joys of I2 itaniums and 8.4 since our OpenVMS platforms have been flagged as "EOL" within the next 2 years or so.)

Ho Hum.

Dave.

[Colin Butcher]

Slightly different, but this recent customer advisory might be relevant:

CUSTOMER ADVISORY: HP Integrity servers blades: Fibre Channel (FC) link fails to come online after upgrade to Emulex HBA FW version 2.02X2 on 456972-B21 8 Gb Emulex mezzanine HBA in the HP Integrity Blade Server configuration. (c04250627) Customer Advisory

Link: http://alerts.hp.com/r?2.1.3KT.2ZR.12iIcE.Kb7o6q..T.fHSo.8UQG.bW89MQ%5f%5fDBeYFQK0

Priority: Recommended

Products: HP Integrity BL Server Blades Release Date: May 08 2014

Description: HBA firmware upgrade on 456972-B21 8Gb Emulex mezzanine HBA results in link going offline when HP 4Gb pass through module is used as interface between HBA and 8Gb FC switch (8 Gb Emulex mezzanine HBA -> 4Gb FC Pass Thru -> HP SN6000  Fibre Channel Switch). As a result of this, the storage behind this HBA will not be accessible.  

Cheers, Colin.