Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Donald L. Gov » Sun, 16 Dec 2001 13:19:38



We are running Sybase 11.9.2 and after upgrading to Solaris 2.8 the
Sybase dataservers seem to go computable for a very long time. The
users complain of very slow responce times. There doesn't seem to be
much or any paging and I/O to the disks is next to nothing. All that
seems to be amiss is the dataservers burning CPU. Has anyone seen
anything like this with Sybase after a 2.6 to 2.8 upgrade?

   Donald L. Gover

 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Anthony Mand » Mon, 17 Dec 2001 10:53:21



Quote:

> We are running Sybase 11.9.2 and after upgrading to Solaris 2.8 the
> Sybase dataservers seem to go computable for a very long time. The
> users complain of very slow responce times. There doesn't seem to be
> much or any paging and I/O to the disks is next to nothing. All that
> seems to be amiss is the dataservers burning CPU. Has anyone seen
> anything like this with Sybase after a 2.6 to 2.8 upgrade?

    Nope. Check the query plan with showplan or sp_showplan.
    However, the query plan shouldn't really change if the Sybase
    ASE version hasn't. Regardless, it wouldn't hurt to check your
    ASE EBF level and your Solaris 8 patches. What else had changed
    with the upgrade? Were the devices affected or the amount of
    shared memory ASE is using?

-am

 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Maj Sweid » Wed, 26 Dec 2001 00:15:41




> > We are running Sybase 11.9.2 and after upgrading to Solaris 2.8 the
> > Sybase dataservers seem to go computable for a very long time. The
> > users complain of very slow responce times. There doesn't seem to be
> > much or any paging and I/O to the disks is next to nothing. All that
> > seems to be amiss is the dataservers burning CPU. Has anyone seen
> > anything like this with Sybase after a 2.6 to 2.8 upgrade?

>     Nope. Check the query plan with showplan or sp_showplan.
>     However, the query plan shouldn't really change if the Sybase
>     ASE version hasn't. Regardless, it wouldn't hurt to check your
>     ASE EBF level and your Solaris 8 patches. What else had changed
>     with the upgrade? Were the devices affected or the amount of
>     shared memory ASE is using?

> -am

Here's something u can look at:
- Check the swap space on the server. Rule of thumb (for any Unix
server with a database engine running on it): Have twice as much swap
space as real memory.
On Sun, u can use these commands to check:
swap -l  ==> List allocated swap space/
swap -s  ==> List free swap space.
If the Sybase is the only thing running on the server, I'd give Sybase
90% of the real memory.

- Write youself a little sproc like sp_who to select "cpu" and
"physical_io" columns from master..sysprocesses. When u see something
slow, run your sproc to see who's got the highest I/Os and CPU usage.
It might be just a coincidence that somebody compiled or running a bad
query out there.

Sorry, Without looking at the server myself, I can't give be more
specific to the cause.

Maj Sweidan

 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Anthony Mand » Thu, 27 Dec 2001 14:24:46



> Here's something u can look at:
> - Check the swap space on the server. Rule of thumb (for any Unix
> server with a database engine running on it): Have twice as much swap
> space as real memory.

     Rubbish. ASE uses shared memory. Swap space isn't really required
     by it. Other processes, however, would be a different issue.

Quote:> On Sun, u can use these commands to check:
> swap -l  ==> List allocated swap space/
> swap -s  ==> List free swap space.
> If the Sybase is the only thing running on the server, I'd give Sybase
> 90% of the real memory.

     Which has what, exactly, to do with swap? You can also configure
     Solaris without swap.

-am

 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Maj Sweid » Fri, 28 Dec 2001 00:10:46




> > Here's something u can look at:
> > - Check the swap space on the server. Rule of thumb (for any Unix
> > server with a database engine running on it): Have twice as much swap
> > space as real memory.

>      Rubbish. ASE uses shared memory. Swap space isn't really required
>      by it. Other processes, however, would be a different issue.

> > On Sun, u can use these commands to check:
> > swap -l  ==> List allocated swap space/
> > swap -s  ==> List free swap space.
> > If the Sybase is the only thing running on the server, I'd give Sybase
> > 90% of the real memory.

>      Which has what, exactly, to do with swap? You can also configure
>      Solaris without swap.

> -am

Sure u can configure Solaris w/o a swap, but how long do u think the
Unix server will keep running before it crashes as a result of having
no swap?!!!
I didn't say swap was the cause to his problem. When Sybase is running
slow, there could be many reasons for the slow response time. If the
DBA can't find anything wrong inside the Sybase server, then he should
take a step back and look at the Unix server. SWAP space is needed
specially when u have a Unix server with many other processes.
 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Maj Sweid » Fri, 28 Dec 2001 01:28:10




> > Here's something u can look at:
> > - Check the swap space on the server. Rule of thumb (for any Unix
> > server with a database engine running on it): Have twice as much swap
> > space as real memory.

>      Rubbish. ASE uses shared memory. Swap space isn't really required
>      by it. Other processes, however, would be a different issue.

> > On Sun, u can use these commands to check:
> > swap -l  ==> List allocated swap space/
> > swap -s  ==> List free swap space.
> > If the Sybase is the only thing running on the server, I'd give Sybase
> > 90% of the real memory.

>      Which has what, exactly, to do with swap? You can also configure
>      Solaris without swap.

> -am

One more thing. I got interrupted while typing my first response.
Sybase DOES use paging space. Do u remember those timeslice errors
Sybase used to have in version 11 and earlier versions?
Some of those errors were caused by lack of swap space. Sybase did fix
most of their timeslice errors, but its use of paging space did not
stop and still part of the core db engine.
As a matter of fact, lack of swap space not only affects performance
of both the Sybase and Unix servers, it could also cause db corruption
on rare occasions under Sybase 11.
Problems related with lack of swap space are not immediate. The
affects usually show up when a server has been up for along time
without rebooting.

Again, Swap space can be looked IF all means were used of searching
for slow response casues inside the Sybase server.

Maj Sweidan

 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Anthony Mand » Fri, 28 Dec 2001 20:45:34



> Sure u can configure Solaris w/o a swap, but how long do u think the
> Unix server will keep running before it crashes as a result of having
> no swap?!!!

     It will run exactly as long as it would with swap.

Quote:> I didn't say swap was the cause to his problem. When Sybase is running
> slow, there could be many reasons for the slow response time. If the
> DBA can't find anything wrong inside the Sybase server, then he should
> take a step back and look at the Unix server. SWAP space is needed
> specially when u have a Unix server with many other processes.

     That depends.

-am

 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Anthony Mand » Fri, 28 Dec 2001 20:49:34



> One more thing. I got interrupted while typing my first response.
> Sybase DOES use paging space. Do u remember those timeslice errors
> Sybase used to have in version 11 and earlier versions?
> Some of those errors were caused by lack of swap space. Sybase did fix
> most of their timeslice errors, but its use of paging space did not
> stop and still part of the core db engine.
> As a matter of fact, lack of swap space not only affects performance
> of both the Sybase and Unix servers, it could also cause db corruption
> on rare occasions under Sybase 11.
> Problems related with lack of swap space are not immediate. The
> affects usually show up when a server has been up for along time
> without rebooting.

     Based on your above comments, it appears you don't really understand
     exactly what swap and virtual memory is under Unix (and in particular
     on Solaris). Running out of paging space means that there is no free
     VM left. If a process crashes as a result of this, its an OS issue.

-am

 
 
 

Sybase Dataservers Using High CPU after Upgrade from solaris 2.6 to 2.8

Post by Scott Howar » Wed, 02 Jan 2002 20:48:58





>>> If the Sybase is the only thing running on the server, I'd give Sybase
>>> 90% of the real memory.
>> (In truth, at 90% there's a good chance you'll end up
>> with the Sybase binaries being paged, which will mean poor performance for
>> Sybase as well).
> A minor point of clarification: since text pages (the executable mappings of
> a process) are read-only, they are sometimes (under *high* memory contention
> conditions, I believe) simply dropped instead of paged -- they can always be

Which is why I said "paged" rather than "paged out" or "swapped". You are
completly correct in that the text pages can be dropped rather than being
paged out, but when they are required again they will be paged in (using
exactly the same process that Solaris uses when you actually run a binary -
it maps the binary to memory, but doesn't actually load it into RAM until
it's actually required).

"Paging" is probably one of the most mis-understood parts of memory
management in Solaris. Paging != swapping as many people think. Paging is
a normal part of Solaris operation - even if you don't have any swap!

  Scott.

 
 
 

1. Sybase Using High CPU after upgrade to 2.8

We have just upgraded to solaris 2.8 and since that time the sybase
users have complained about poor responce time. When looking that them
system TOP and Sar both show that there's is much higher CPU usage by
the Sybase dataservers then before the upgrade. Has anyone seen this
before? Is this just a side effect of 2.8 being bigger and slower or
is there a problem with Sybase in this environment. We Currently are
running Sybase 11.9.2.

   Thanks in Advance.

   Donald L. Gover

2. Fsck problem on (SCO) Unixware

3. Tips on upgrading Solaris 2.6 to Solaris 2.8

4. RAW mkLinux DR3

5. Upgrade Solaris 2.6 -> 2.8

6. UDB memory

7. Tip on Upgrading solaris 2.6 to 2.8

8. slackware 4

9. Solaris 2.6 to 2.8 Upgrade Failure

10. How do I upgrade solaris 2.6 to 2.8 over serial connection?

11. Upgrade of Solaris 2.6 to 2.8

12. Jumpstart for Solaris 2.5.1 versus Solaris 2.6, 2.7, 2.8

13. solaris 2.6 to solaris 2.8