[AccessD] [dba-VB] [dba-SQLServer] HELP, server completely unresponsive

jwcolby jwcolby at colbyconsulting.com
Mon Sep 28 15:44:34 CDT 2009


Drew,

 >So if you have three cores cranking away, not specifically with calculating on their own, but in 
running things through the busses, then that fourth core may be 'available', but the line is going 
to be busy trying to get to it.  Does that make sense?

No, it doesn't work that way.  Yes, all four cores use the same busses but each has an equal shot at 
  everything - shared cache, IO, busses etc.

 > Plus, drive operations usually take a very high priority when it comes to interrupt requests, 
that is why almost any machine, when it's processor is tied up with drive operations, is going to 
appeared locked.

This may in fact be the case.  MS doesn't specifically say.  However what is "supposed" to happen is 
that low impact / low IO stuff is supposed to get priority exactly because it has a low impact and 
costs nothing to process.  Hi impact stuff is supposed to get last priority exactly to prevent 
locking out the low impact stuff.  If a uart is running bringing in a few characters every 
millisecond then it is supposed to be the highest priority so that its pitiful little I/O is 
processed.  Like that.  The high impact stuff is supposed to grab all of the remaining bandwidth / 
processor cycles after the puny stuff is done.

More or less.

 > #2.  RAID 6.  Let's face it, the absolute best method for performance AND reliability is a RAID 
10 (or a RAID 0 +1).

No argument.  Every RAID class has it's purpose.  Raid 6 allows two drives to fail without data 
corruption, yet only uses two drives for parity (redundancy).  Raid 10 uses 1/2 of your drives for 
redundancy.

 >But RAID 10 is the most expensive to implement, however, with the cost of today's drives, that 
really should not be used as a factor in decision making here, unless you are going with ultra huge 
and ultra expensive drives!

That is easy to say when you are working for a corporation and just put in a funds request.  My 
funds requests come directly out of my children's mouths.

 > #3.  This is just guessing, but you probably also have your OS on a partition on the same RAID.

Nope, I have Raid1 for the OS.  I have the log files out on it's own RAID array (separate disk set).

  Solution #1.  Upgrade your machines.  Put in 2 separate (duo or quad core) processors.  Mirror two 
drives for the OS, and then 4 more drives for your data RAID 10.

Please do put in a funds request and send the check to me at ... ;)

 > Solution #2.  Upgrade to Windows 2008, and use it without the GUI (2008 can be setup where it's 
just a command line interface, and thus it strips all the regular Window's overhead out of the 
mix....) and then access it strictly through the network from another computer (using Enterprise 
manager).

Uhh... I don't think so.

I should be able to see the SQL Server instance from another machine.  It is not showing up (when 
whatever this is happens), most likely because the machine is not responding before the remote 
Management studio times out.

I should be able to remote desktop into the machine as it is.  The machine is "not responding".  It 
seems unlikely that I am gonna change that with an OS change and stripping the gui.  A lot of work 
with a highly questionable result set.

As it happens I do have Server 2008 and SQL Server 2008 as well.  Given that my data is all on raid 
disks, I can fairly easily swap out the Server OS and SQL Server instance.  "fairly easily" being of 
course many hours of my own time (no corporate IT department to call here).

I will eventually do this just because I need to do so to stay modern.  However I certainly do not 
expect that to solve this problem.  MS has had a dozen years to stop this nonsense and hasn't.  I 
have no expectation that they suddenly decided fixing bugs is more important that sales brochures 
bullet points.

John W. Colby
www.ColbyConsulting.com


Drew Wutka wrote:
> Ok, I had to do a little hunting, to find your system configuration.
> 
> You have 2 Windows 2003 x64 servers.  Both with a quad core processor,
> 16 gigs of RAM and a RAID 6.
> 
> Ok, don't take this as the word of God or anything, but there are a few
> issues that are probably hitting you:
> 
> #1.  A quad core processor is not the same as having 4 actual
> processors.  It's similar.  You just have 4 cores.  The biggest
> difference is that with a quadcore (or duocore) system, all of the
> 'processors/cores' are using the same architecture/busses to communicate
> with everything else.  So if you have three cores cranking away, not
> specifically with calculating on their own, but in running things
> through the busses, then that fourth core may be 'available', but the
> line is going to be busy trying to get to it.  Does that make sense?
> Plus, drive operations usually take a very high priority when it comes
> to interrupt requests, that is why almost any machine, when it's
> processor is tied up with drive operations, is going to appeared locked.
> 
> #2.  RAID 6.  Let's face it, the absolute best method for performance
> AND reliability is a RAID 10 (or a RAID 0 +1).  That's a striped set
> that is mirrored.  You get DOUBLE the reading spead of the already
> doubled striped set (cause it can read data from all four drives (in a 2
> drive striped (and then 2 mirror drives) config)) at the same time.  And
> you get the normal double write speed.  But RAID 10 is the most
> expensive to implement, however, with the cost of today's drives, that
> really should not be used as a factor in decision making here, unless
> you are going with ultra huge and ultra expensive drives!  RAID 5 is the
> cheap way to get redundancy, and RAID 6 is a little more 'expensive' to
> get double failure capability in a RAID set.  DROP THE RAID 6, and go
> with a STRIPED MIRROR!
> 
> #3.  This is just guessing, but you probably also have your OS on a
> partition on the same RAID.  For a desktop, having a good RAID 10, it's
> gonna be fun trying to get 6 disks in there (4 for the striped mirror
> data RAID, and 2 for a plain mirrored OS drive).  But if your OS is on
> the same RAID set as your data, you are going to be getting a
> performance hit.
> 
> So, what I would recommend is one of two things:
> 
> Solution #1.  Upgrade your machines.  Put in 2 separate (duo or quad
> core) processors.  Mirror two drives for the OS, and then 4 more drives
> for your data RAID 10.
> 
> Solution #2.  Upgrade to Windows 2008, and use it without the GUI (2008
> can be setup where it's just a command line interface, and thus it
> strips all the regular Window's overhead out of the mix....) and then
> access it strictly through the network from another computer (using
> Enterprise manager).
> 
> Drew




More information about the AccessD mailing list