<div dir="ltr"><font>
</font><div class="" lang="x-western">
<font face="Courier New, Courier, monospace">Hi,<br>
I came across this paper ...<br>
<br>
</font><font face="Courier New, Courier, monospace"><br></font><pre><a class="" href="http://www.ece.rice.edu/%7Epjv/mclock.pdf">http://www.ece.rice.edu/~pjv/mclock.pdf</a></pre><font face="Courier New, Courier, monospace">
<br>
while going thru a VMWare blog and this is the brief about what
it claims to be :<br>
<br>
<i>In this paper, we presented a novel IO scheduling algorithm,</i><i><br>
</i><i>mClock, that provides per-VM quality of service in
presence of variable overall throughput. </i><i><br>
</i><i>The QoS re-quirements for a VM are expressed as a minimum
reservation, </i><i>a maximum limit, and a proportional
share. </i><i><br>
</i><i><br>
</i><i>A key aspect of mClock is its ability to enforce such
controls even with fluctuating overall capacity, as shown by</i><i>
</i><i>our implementation in the VMware ESX server hypervisor. </i><i><br>
</i><i><br>
</i><b><i>We also presented dmClock, a distributed version of</i></b><b><i><br>
</i></b><b><i>our algorithm that can be used in clustered
storage system architectures. </i></b><b><i><br>
</i></b><b><i>We implemented dmClock in a distributed storage
environment and showed </i></b><b><i><br>
</i></b><b><i>that it works as specified, maintaining global
per-client reservations,</i></b><b><i><br>
</i></b><b><i>limits, and proportional shares, even though the
schedulers run locally </i></b><b><i><br>
</i></b><b><i>on the storage nodes.</i></b><b><br>
<br>
</b>I thought this was worth sharing with the wider audience to
see if there is any value
in looking at dmclock from a GlusterFS perspective ?<br>
<br>
thanx,<br>
deepak<br>
</font></div><font><br></font></div>