It would help in analysing if you could attach the outputs of profile info etc as files and we can probably dig deep into why you are observing what you are observing. Pranith ----- Original Message ----- > From: "Kal Black" <kaloblak at gmail.com> > To: "Pranith Kumar Karampuri" <pkarampu at redhat.com> > Cc: "Kaleb S. KEITHLEY" <kkeithle at redhat.com>, gluster-users at gluster.org > Sent: Wednesday, September 11, 2013 11:40:32 PM > Subject: Re: Native client reads from a replication volume > > Thank you for the information Guys .. it is really helpful !!! > Is that means that what I am observing is indeed weird or is expected with > small files? > Just for the sake of curiosity, if I assume that what I am seeing is due to > the small file sizes, do you have any idea or suggestion what file size up > would trigger the expected behavior so I can repeat my test using this kind > of files? > > Again .. very much appreciated! > Kal > > > On Wed, Sep 11, 2013 at 2:01 PM, Pranith Kumar Karampuri < > pkarampu at redhat.com> wrote: > > > > > > > > > The client sends a LOOKUP to all bricks. All servers will respond. The > > > first response the client gets dictates which server it will send > > > subsequent STAT, READ, etc., calls to. > > > > > To add more clarity: After both servers respond to the LOOKUP, if both of > > them are good copies it will choose whichever responded first. If the > > mount is on same machine it sends subsequent STAT/READ etc to local brick. > > Otherwise the one which has the good copy is chosen as the "read-child" and > > all subsequent requests like STAT, READ will be sent to that. > > > > Pranith. > > >