<div dir="ltr"><div><div><div><div><div>Hi,<br></div>we have recentily upgraded the swift cluster hosting images for wikimedia from folsom to icehouse and it has been very smooth! thank you!<br><br></div>After the upgrade we have noticed a substantial increase in the bandwidth used by backends, in particular it seems to line up with the upgrade of frontends to icehouse (which happened last).<br>

</div>You can see the effect I'm talking about here together with the upgrade timeline: <a href="https://wikitech.wikimedia.org/wiki/Swift/Icehouse#post-upgrade">https://wikitech.wikimedia.org/wiki/Swift/Icehouse#post-upgrade</a><br>

<br></div>It seems to coincide with an increase in CPU on the backends too: <a href="http://ganglia.wikimedia.org/latest/?r=custom&cs=07%2F07%2F2014+00%3A00+&ce=07%2F21%2F2014+00%3A00+&m=cpu_report&s=by+name&c=Swift+eqiad&h=&host_regex=&max_graphs=0&tab=m&vn=&hide-hf=false&sh=1&z=small&hc=4">http://ganglia.wikimedia.org/latest/?r=custom&cs=07%2F07%2F2014+00%3A00+&ce=07%2F21%2F2014+00%3A00+&m=cpu_report&s=by+name&c=Swift+eqiad&h=&host_regex=&max_graphs=0&tab=m&vn=&hide-hf=false&sh=1&z=small&hc=4</a><br>

<br></div>I was wondering what has changed that could cause that increase and if it is expected, also what would be a good way to graph replication bandwidth vs "serving" bandwidth? that'd be useful too I think.<br>

<br>thanks in advance!<br>filippo<br></div>