<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Fri, Feb 7, 2014 at 1:25 PM, Alex Samad - Yieldbroker <span dir="ltr"><<a href="mailto:Alex.Samad@yieldbroker.com" target="_blank">Alex.Samad@yieldbroker.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi<br>
<br>
Q) am I being unrealistic to think I should be able to get 10Gb/s routing/firewall in a vm?</blockquote><div><br></div><div>If you mean 10Gbps full duplex (tx+rx) with real-world packet sizes: yes. very unrealistic. not even close.</div>
<div>If you mean 10Gbps with jumbo frames, small # of prefixes then yes its possible.<br></div><div>If you want it to be capable of standing up under load of a DDoS attack of 64 byte frames: you're dreaming.</div><div>
<br>(linerate 64-byte 10G is 14.88M PPS.)</div><div><br></div><div>The "best" you can do on x86 hardware is around 600-800K PPS per 'core' of most modern Intel Xeon 56xx and a multi-queue-capable NIC. So maybe 2-3M PPS best case.</div>
<div>With no features other than forwarding.</div><div><br></div><div>Add anything interesting like sFlow, ACLs, policing/shaping etc and it drops dramatically.</div><div><br></div><div>The unfortunate reality is that there is about 4 or 5 orders of magnitude difference in performance between what 'software' on a general purpose CPU can do and what dedicated network silicon can do for an "equvalent price" silicon.</div>
<div><br></div><div><br></div></div></div></div>