<div dir="ltr"><div class="gmail_extra"><br><div class="gmail_quote">On 1 October 2015 at 01:38, Tom Pepper <span dir="ltr"><<a href="mailto:tom@encoding.com" target="_blank">tom@encoding.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div id=":2jz" class="a3s" style="overflow:hidden">If you can stomach a small recovery window, I'd instead suggest vanilla NFS, snapshots, and zfs send/receive to keep a nearline replica. You'll have much higher sequential I/O if so.<br></div></blockquote><div><br><br></div><div>You mean not use glusters/ceph at all? a pure zfs solution? I have considered a setup like that, it has its advantages and we could wear the recovery window. But it does require manual maintenance and switch over when a node goes down as opposed to the automation of gluster. But the performance is very good.<br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div id=":2jz" class="a3s" style="overflow:hidden">
<br>
LACP will also restrict single socket performance to 1Gbps, but multi socket can exceed of course.</div></blockquote><div><br></div><div>Its a PITA that when dealing with two/three nodes setups :( <br><br><br></div><div>thanks.<br><br></div></div><br><br><br clear="all"><br>-- <br><div class="gmail_signature">Lindsay</div>
</div></div>