Re: (Java) Synchronizing multiple app servers through Oracle

On Dec 21, 5:19 am, "Palooka" <nob..._at_nowhere.com> wrote:
> "joel garry" <joel-ga..._at_home.com> wrote in message
>
> news:a21025c6-99a0-4049-a975-188c77bbef28_at_d27g2000prf.googlegroups.com...
>
> > On Dec 19, 12:05 pm, chrism..._at_gmail.com wrote:
> >> > > Typically the lock will last less than one second.
>
> >> > I'd say that's pretty long. How often will you - pardon, your
> >> > application - acquire the lock and what's the general load pattern of
> >> > the application?
>
> >> Yes, it's not that short of a period of time, but unfortunately,
> >> that's how long it takes to grab unique records. Right now we have 2
> >> app servers each hitting the database once per second and acquiring a
> >> lock. However, the number of app servers could grow to 10 or more. I
> >> wrote a multi-threaded test to simulate multiple servers, and found
> >> that Oracle appears to have a problem when more than 10 locks are
> >> acquired at the same time. It seems to just give up and wait for the
> >> timeout once too many have been acquired. Is there a setting in
> >> Oracle to handle more simultaneous locks?
>
> > It kind of depends what exactly is going on. Metalink Note:15476.1 is
> > a basic description of locks and how to see what is locked. Note:
> > 402983.1 is a performance FAQ that includes how to check if the db is
> > really hung. The Performance manual in the doc set describes how to
> > check what is really being waited for. Views such as dba_lock and v
> > $session_wait can really help, and don't discount some of the graphic
> > tools available.
>
> > If you are not using the default locking methods, you need to
> > understand what you are doing and how to use the instrumentation
> > available to understand it. Once you understand the actual problem,
> > you can fix it, if you haven't designed your strategy to be
> > unfixable. Asking for a setting for more simultaneous locks - well,
> > you might want to review the Concepts manual on locking, you may have
> > some misconceptions.
>
> > Before you do any of that... when "oracle appears to have a
> > problem..." are your cpu's going nuts? Do you see very high latch
> > waits? What version and platform, exactly (10g isn't enough, we need
> > something like 10.2.0.4), are you using? What locks are you aquiring?
>
> Joel is rightly suspicious that the wheel is being reinvented.

Nitpick: I think he talks less about reinventing the wheel but rather about understanding concepts and what happens and apply the correct medicine - not something that you guess will help. :-)

Kind regards

robert Received on Fri Dec 21 2007 - 03:35:10 CST