Hyperthreading killing performance of simple query?

We have a new server, 4 physical dual core processors, with hyperthreading,
so it shows up as 16 logical processors.
A simple query joining 3 tables, a distinct, an order by, with maxdop set to
0, takes over 2 seconds. With maxdop set to 1, it takes 200 milliseconds.
setting maxdop between 1 and 15, pretty linear time progression.
Does this sound sane? Is there some magic matrix where under load a higher
maxdop will result in this simple query running faster than where there is
no load (the current case)
Thanks in advance!
DerrickDerrick,
It is true: Parallelism does not alway help. In fact, for some queries it
may make them perform abysmally.
Although I do not do this personally, several posters recommend setting an
OLTP server to a maximum degree of parallelism of 1. They indicate good
success and few discernable problems because of the setting.
I do follow the guideline to limit the server's maximum degree of
parallelism to the number of physical chips, which in your case would be 4.
Then, if a query is troubled we can specifically run it at MAXDOP 1.
FWIW,
RLF
"Derrick" <derrick1298@.excite.com> wrote in message
news:%23E%23LO250HHA.5644@.TK2MSFTNGP05.phx.gbl...
> We have a new server, 4 physical dual core processors, with
> hyperthreading, so it shows up as 16 logical processors.
> A simple query joining 3 tables, a distinct, an order by, with maxdop set
> to 0, takes over 2 seconds. With maxdop set to 1, it takes 200
> milliseconds. setting maxdop between 1 and 15, pretty linear time
> progression.
> Does this sound sane? Is there some magic matrix where under load a
> higher maxdop will result in this simple query running faster than where
> there is no load (the current case)
> Thanks in advance!
> Derrick
>|||Russell,
Thanks for the info, one more q, so would you reccomend hyperthreading be
enabled, or disabled?
We ran tests both ways and turning hyperthreading off does not seem the way
to go based on our results, does that match up with what you have seen/read?
Thanks again,
Derrick
"Russell Fields" <russellfields@.nomail.com> wrote in message
news:eJddSQ60HHA.3788@.TK2MSFTNGP02.phx.gbl...
> Derrick,
> It is true: Parallelism does not alway help. In fact, for some queries it
> may make them perform abysmally.
> Although I do not do this personally, several posters recommend setting an
> OLTP server to a maximum degree of parallelism of 1. They indicate good
> success and few discernable problems because of the setting.
> I do follow the guideline to limit the server's maximum degree of
> parallelism to the number of physical chips, which in your case would be
> 4. Then, if a query is troubled we can specifically run it at MAXDOP 1.
> FWIW,
> RLF
> "Derrick" <derrick1298@.excite.com> wrote in message
> news:%23E%23LO250HHA.5644@.TK2MSFTNGP05.phx.gbl...
>|||Derrick,
Lara Rubbelke made these comments:
http://blogs.digineer.com/blogs/lar...rthreading.aspx
He links to Slava Ok's
http://blogs.msdn.com/slavao/archiv.../12/492119.aspx
Their bottom seems to be, do not enable hyperthreading unless you have
proved it will help you.
RLF
"Derrick" <derrick1298@.excite.com> wrote in message
news:u1F0WU60HHA.464@.TK2MSFTNGP02.phx.gbl...
> Russell,
> Thanks for the info, one more q, so would you reccomend hyperthreading be
> enabled, or disabled?
> We ran tests both ways and turning hyperthreading off does not seem the
> way to go based on our results, does that match up with what you have
> seen/read?
> Thanks again,
> Derrick
>
> "Russell Fields" <russellfields@.nomail.com> wrote in message
> news:eJddSQ60HHA.3788@.TK2MSFTNGP02.phx.gbl...
>