port queue depth - the never ending story.

I am starting to hate going overseas for a holiday.  I always end up on deaths door step and this time nothing changed.  So, I now have a week off work to try to get over this thing I got in Thailand and it is not attributed to the local inhabitants.  Only once before in my life have I been so sick.  The good side is that I have lost a few kilo's.  Surgical masks should be compulsory on all international flights.

I have been watching the HDS forums (between coughing and sleeping and my big screen TV wont power on) and one discussion is about port queue depths.  I raised this issue with Pierre Raymond of HDS just over a month ago (as part of performance management with HDS products) and he got extremely excited about this stating that is was impossible (in a French Canadian accent) to fill up the port queue depths on a USP and if you were really stupid, you could cause problems on a 9980V.  He said it was not possible.  No Way.  Can't do it.  Zip.

I wish he had explained his knowledge about this as it seems VMware can quickly fill up a port.  Going by the calculations with ESX Server, you can only have 8 LUNs on a 9980V and thus 16 on a USP (V).  Thats a port queue depth of 64 per LUN and only 512 available on the 9980V.

My glory days of being a Solaris guru was spent with days working out if I needed to adjust sd_max_throttle and no, it was never done but it made my boss happy to think I knew what I was doing.

So why is there so much discussion about this but never anything concrete from the vendors that states exactly how this can be looked at.  Finisar Netwisdom (I just have to get that one day) is capable of monitoring this and alarming if the port on the storage goes over the defined setting.  Does that mean that port queue depths are all happening at different times?  So I could be having 500 from host A and 500 from host B and 500 from host C, etc, etc but never at the same time? 

I have five heavily utilised Exchange servers on one port on our USP (clusters 1 and 2).  They each have the standard settings of 16 but these should be changed to 32 going by MS's recommendation.  Each server has 10 LUN's.  I am currently using about 800 of the available 1024 port queue depths.  If I changed all the HBA settings to 32, then would this means the servers start to get poor performance?  800 x 2 = 1600 which is definitely much more than 1024.  I watch those ports like a hawk just to see what goes on as any problem with Exchange causes thousands of calls to our Service Desk.

We are starting to put in some bleeding edge networked server virtualisation software so I am eagerly waiting to see what I have to do for this in the way of storage.  Can any one out there state if we really need to worry about queue depths considering Pierre's remark on this?

Does anyone really care?