Long Spins
A user asked how to reduce "long spins" reported by IDS and how they relate to the number of virtual processors. Replies cited the manual's advice (fewer VPs, less machine load, NOAGE or processor affinity), and explained that spin locks protect objects across processes, so configuring more CPU VPs than physical CPUs (or restricting VPs to fewer cores via affinity) causes spinning while the OS reschedules the lock holder. Art Kagel cautioned that cutting CPU VPs may just slow things down and asked for onstat output (-c, -p, -g glo/iov/iof/wai/rea/ses) to find the real contention; the poster never supplied it, so no resolution is recorded.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
Hi, I want to know how we can reduce longspins ? What is the relation of longspins with the number of VP ? Regards
Hi, The IBM IDS document says: "To reduce longspins, reduce the number of virtual processors, reduce the load on the computer, or, on some platforms, use the no-age or processor affinity features." But wait for other to advice you, I'm sure the IDS guru's will suggest you a solution if one is really needed. Pls mention the IDS version and OS, that realy helps ! Regards, Vikas ***************************************************************************** Hi, I want to know how we can reduce longspins ? What is the relation of longspins with the number of VP ? Regards
Why do you think that you have a problem with long spins? Art Art S. Kagel Oninit (www.oninit.com) IIUG Board of Directors (art@iiug.org) Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on my employer, Oninit, the IIUG, nor any other organization with which I am associated either explicitly or implicitly. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Mon, May 25, 2009 at 5:57 AM, SHAHZAD KASI <skasi@i2cinc.com> wrote: > Hi, > I want to know how we can reduce longspins ? What is the relation of > longspins > with the number of VP ? > > Regards > > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --001636c5a49077bd27046abd1624
Hello, I see that there are many threads waiting for the same resources, thus, it increases long spins. I also would like to know that how should i reduce it, and also to know, how can i trace the users who are yielding long spins. Thanks
Hello, Thanks for the valuable answer. I would like to ask how reducing the number of processors will decrease the Long spins? Is there any particular ratios for the number of VP for different classes e.g. aio, cpu, tli, etc, where we can improve the overall performance of the system.
Post the following:
onstat -c
onstat -p
onstat -g glo
onstat -g iov
onstat -g iof
onstat -g wai
onstat -g rea
onstat -g ses
Elapsed time since onstat -z was last run (or when the engine was started if
onstat -z hasn't been run since startup).
Art
Art S. Kagel
Oninit (www.oninit.com)
IIUG Board of Directors (art@iiug.org)
Disclaimer: Please keep in mind that my own opinions are my own opinions and
do not reflect on my employer, Oninit, the IIUG, nor any other organization
with which I am associated either explicitly or implicitly. Neither do
those opinions reflect those of other individuals affiliated with any entity
with which I am affiliated nor those of the entities themselves.
On Tue, May 26, 2009 at 7:24 AM, SHAHZAD KASI <skasi@i2cinc.com> wrote:
> Hello,
> I see that there are many threads waiting for the same resources, thus, it
> increases long spins. I also would like to know that how should i reduce
> it,
> and also to know, how can i trace the users who are yielding long spins.
>
> Thanks
>
>
>
>
*******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>
--001636c5a87e4dc67a046ad07bfd
Reducing the number of CPU VPs MAY or MAY NOT decrease longspins. It may
simply slow things down considerably. What you have to do is to determine
where the contention for resources that is causing the spins lay and
increase those critical resources until the contention is minimized. Post
the onstat output and other information I requested and we'll see if we can
spot the problem.
Art
Art S. Kagel
Oninit (www.oninit.com)
IIUG Board of Directors (art@iiug.org)
Disclaimer: Please keep in mind that my own opinions are my own opinions and
do not reflect on my employer, Oninit, the IIUG, nor any other organization
with which I am associated either explicitly or implicitly. Neither do
those opinions reflect those of other individuals affiliated with any entity
with which I am affiliated nor those of the entities themselves.
On Tue, May 26, 2009 at 8:57 AM, SHAHZAD KASI <skasi@i2cinc.com> wrote:
> Hello,
> Thanks for the valuable answer. I would like to ask how reducing the number
> of
> processors will decrease the Long spins? Is there any particular ratios for
> the number of VP for different classes e.g. aio, cpu, tli, etc, where we
> can
> improve the overall performance of the system.
>
>
>
>
*******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>
--001636c923d7580bae046ad083b0
The purpose of a spin lock is to lock the object amongst multiple processes. The mutex is to lock threads from each other. If you have configured more cpuvps than physical cpus, then you can encounter lock inversion which can only be resolved by spinning until t= he process holding the spin lock is scheduled by the OS. = "SHAHZAD KASI" = <skasi@i2cinc.com = > = To Sent by: ids@iiug.org = ids-bounces@iiug. = cc org = Subj= ect Re: Long Spins [15833] = 05/26/2009 07:57 = AM = = = Please respond to = ids@iiug.org = = = Hello, Thanks for the valuable answer. I would like to ask how reducing the nu= mber of processors will decrease the Long spins? Is there any particular ratios= for the number of VP for different classes e.g. aio, cpu, tli, etc, where w= e can improve the overall performance of the system. ***********************************************************************= ******** Forum Note: Use "Reply" to post a response in the discussion forum. =
Similarly if processor affinity has limited the CPU VPs to a smaller number of physical processor cores than the number of VPs. Art Art S. Kagel Oninit (www.oninit.com) IIUG Board of Directors (art@iiug.org) Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on my employer, Oninit, the IIUG, nor any other organization with which I am associated either explicitly or implicitly. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 26, 2009 at 10:09 AM, Madison Pruet <mpruet@us.ibm.com> wrote: > The purpose of a spin lock is to lock the object amongst multiple > processes. The mutex is to lock threads from each other. > > If you have configured more cpuvps than physical cpus, then you can > encounter lock inversion which can only be resolved by spinning until t= > he > process holding the spin lock is scheduled by the OS. > > = > > "SHAHZAD KASI" = > > <skasi@i2cinc.com = > > > = > To > > Sent by: ids@iiug.org = > > ids-bounces@iiug. = > cc > > org = > > Subj= > ect > > Re: Long Spins [15833] = > > 05/26/2009 07:57 = > > AM = > > = > > = > > Please respond to = > > ids@iiug.org = > > = > > = > > Hello, > Thanks for the valuable answer. I would like to ask how reducing the nu= > mber > of > processors will decrease the Long spins? Is there any particular ratios= > for > > the number of VP for different classes e.g. aio, cpu, tli, etc, where w= > e > can > improve the overall performance of the system. > > ***********************************************************************= > ******** > > Forum Note: Use "Reply" to post a response in the discussion forum. > > = > > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --001636c5ab406c140c046ad160c0