Re: received more blocks than served?

From: Dan Norris <dannorris_at_dannorris.com>
Date: Thu, 10 Jul 2008 12:58:22 -0600
Message-ID: <48765BCE.3010009@dannorris.com>

Yong,

Here's the scenario why my counting method didn't work:

When all instances start at the same time:
Inst1: send 500, recv 200
Inst2: send 300, recv 400
inst3: send 200, recv 400

Total: send 1000, recv 1000

Now, at some later time, inst1 is restarted so it's counters are now 0. At that point, the totals are send: 500, recv: 800. It's that scenario where you will see recv > send. That fits my case since one of my instances was restarted relatively recently, but the other two have were started at the same time (a long time ago).

Hope that clears things up.

Dan

Yong Huang wrote:

Date: Tue, 08 Jul 2008 11:21:32 -0600
From: Dan Norris <dannorris@dannorris.com>
Subject: Re: received more blocks than served?

 I think Mark's suggestion about the different start times is probably the
culprit. Flawed logic on my part--thanks Mark for pointing it out. Guess I
was having a Monday :-|
Dan
    

Dan and Mark,

I still don't understand. That seems to say that when RAC instances are started
at different times, some blocks will be received from nowhere, or (as I see on
some of my RAC nodes) some blocks will be sent to nowhere. Our current thinking
is the cluster-wide total count by summing on gv$ views should still be
balanced, taking into account other stats such as lost or corrupt blocks,
blocks converted from current to CR, which are generally very small or 0. There
must be something we're missing.

If the blocks are in transit between instances, the total count will not
balance. But your query shows a big difference (SERVED - RECEIVED). I don't
think we can explain that big number by blocks in transit.

Yong Huang


      

  

-- http://www.freelists.org/webpage/oracle-l Received on Thu Jul 10 2008 - 13:58:22 CDT