In the course of digging for some info on IPv6 stack tuning, I discovered an interesting comment that I thought I'd share, regarding IPv6 performance on BSD (i.e. Mac OSX) added to the very bottom of an article I've referenced previously.
quote:Testing has shown that on end-to-end 10G paths, IPV6 appears to be about 40% slower than IPV4 on FreeBSD 7.3, and 20% slower on FreeBSD 8.2. This is a known FreeBSD issue, and will be addressed in a future release.
There had been some previous discussion about the visible downstream performance downgrade users were seeing when testing with Comcast's IPv6 speed test which I have experienced myself, as well, as demonstrated in this screenshot from my Macbook running OSX 10.9 Mavericks from my own test of my U-verse connection rated at 45/6Mbps.
[att=1]
In any case, IPv6 natively adds another 20 Bytes to the protocol header, so it has 50% more overhead per packet than IPv4 at 40 Bytes. With AT&T there is also an additional 28 Bytes of overhead for the 6rd tunnel encapsulation, as well. So in 6rd tunnel mode each IPv6 packet will have a minimum of 88 bytes of overhead versus an IPv4 packet that natively has 40 Bytes plus, in the case of my Mac, another 12 Bytes of TCP Timestamp overhead. That is a 41% increase in overhead per tunneled IPv6 packet. On transactions using much smaller payloads than the MSS, this could be a noticeable increase in total overhead. I found an article discussing test results on IPv6 overhead and in one case it shows a 55.6% increase on return traffic.
The added overhead should not have this noticeable of an impact on the overall throughput, though. Best case with IPv4 on a fully loaded packet you will have ~3.5% overhead. With IPv6 over 6rd, the best case would be ~5.9% overhead. TCP should be able to tune the MSS up to 1412 Bytes per packet with a fairly large window size. So the 45% reduction in IPv6 throughput performance, as compared to IPv4, at this point, I can only attribute to whatever the specific underlying OS issue is with BSD.
--
Scott, CCIE #14618 Routing & Switching
http://rolande.wordpress.com/
↧