RAC Re: Global Cache Null to X wait event..

  • From: "Dubovecky, Jeffrey P" <jpdubovecky@xxxxxxxxxx>
  • To: <racdba@xxxxxxxxxxxxx>
  • Date: Wed, 23 Apr 2008 13:44:33 -0400


Here are the statspack reports for both instances of the 2 node cluster
(Oracle 9i/AIX O/S).  The network admins are having a hard time
understanding that having the interconnect on the public network is an
issue considering that they don't see a bandwidth issue (although
they've seen some UDP fragmentation).  A question I have is that if we
have UDP fragmentation, how does Oracle RAC recover from a dropped/lost
UDP packet?

The other piece I still wanted some assistance on was how to avoid the
Null to X wait event outside the scope of the interconnect issue.  Our
process right now basically inserts records into two sets of tables, one
set for maintaining current state, and one set for staging that gets
copied over to another database approximately 1-2 minutes afterwards via
user job in 9i which could go to either instance in the cluster.  Once
the data is copied over to the other database, the records get removed
from these staging tables, and this is we're we've seeing the Global
Cache Null to X wait event.

Basically, we've been doing application partitioning (one set of common
tasks connected to one instance) and I've been looking at different
types of alternatives (setting PCTFREE high, reverse key indexes) as to
avoid the possibility of solving the inter-instance issue, however if we
have a good interconnect setting, wouldn't this solve the majority of
the issues we're seeing?

Thanks JPD

-----Original Message-----
From: racdba-bounce@xxxxxxxxxxxxx [mailto:racdba-bounce@xxxxxxxxxxxxx]
On Behalf Of K Gopalakrishnan
Sent: Wednesday, April 23, 2008 10:21 AM
To: racdba@xxxxxxxxxxxxx
Subject: RAC Re: Global Cache Null to X wait event..


NULL-X is a wait during the new lock/block acquisition and there will be
numerous messages sent/received between the master and requestor (and
may be holder if your cluster is 3node and above). When you use the
public network for messaging traffic, you will not only exaggerate the
waits for NULL-X wait, but also the waits for ALL GC related work.

Do you care to share the  statspack/AWR?


On Wed, Apr 23, 2008 at 6:59 AM, Dubovecky, Jeffrey P
<jpdubovecky@xxxxxxxxxx> wrote:
> RAC DBA's,
> I'm running into an issue with a Global Cache Null to X wait event 
> that we've found during an LTOM trace.  It seems that the issue in 
> question is that we're inserting records into a block and deleteing 
> different records from the same block, or at least that's how it
appears from the LTOM trace.
> We also know we have interconnect issues, as the interconnect is on a 
> public network and not on a private network.  This wait event happens 
> a few times during the day, however it can take a PL/SQL routine up to

> 20 seconds to complete and is not specific to a specific PL/SQL 
> routine that may insert/delete records from the same set of tables.  
> So here are my
> questions:
> 1.  Could the interconnect being on a public network exaggerate the 
> Global Cache Null to X wait event?
> 2.  Besides rewriting the application to perform the delete at a later

> timeframe, what other suggestions can be implemented to reduce this 
> wait event.
> Thanks for any assistance you can provide me in advance - JPD
> Jeff Dubovecky
> Consulting Engineer
> Union Switch & Signal
> 1000 Technology Drive
> Pittsburgh, PA  15219-3120
> Ph: 412-688-2881
> Fax: 412-688-2399
> Email:  jpdubovecky@xxxxxxxxxx

Best Regards,
K Gopalakrishnan
Co-Author: Oracle Wait Interface, Oracle Press 2004

Author: Oracle Database 10g RAC Handbook, Oracle Press 2006

Other related posts: