55

Are there any practical speed/latency gains from using IPv6, or is it just to make room for more IPs?

9
  • 6
    there's other advantages though, like rendering NAT practically obsolete, and allowing pretty much anyone to have a static ip address
    – Journeyman Geek
    Aug 28, 2010 at 1:36
  • 1
    Well I dunno if NAT is "obsolete", maybe unnecessary, but I could see benefits in having your network secured off from the rest of the world save for some basic entry points. Also I think most anyone who wants a static IP currently can have one, ISPs just find it easier to assign them dynamically. You still end up with the same amount of IPs given out.
    – Jane Panda
    Aug 28, 2010 at 17:49
  • 3
    @Bob Any real security (obscurity does not count) you got from NAT can be done in IPv6 with a stateful firewall.
    – Azendale
    Jan 1, 2012 at 4:08
  • 1
    And yes, Azendale is exactly right. Any security benefit from NAT you can also get with a stateful firewall, which is basically what NAT is doing in order to accomplish its routing anyway. And you get the added "obscurity" benefit that anyone who has a /64 has 18,446,744,073,709,551,615 possible IP addresses to choose from. Jul 22, 2013 at 5:10
  • 1
    @Bob When looking for a new ISP just last month, one question I specifically asked was whether they provide for the possibility of statically assigned IP addresses. Out of five or six ISPs I contacted, one answered that they offer that option (at an extra monthly cost), and the one that did offer statically assigned IP addresses certainly is not one of the major/mainstream ISPs. I had never heard of them before. I'd say that ISPs providing statically assigned IP addresses on consumer connections, even at a cost, is definitely the exception rather than the rule.
    – user
    Nov 15, 2013 at 14:32

6 Answers 6

37

Practical impact? Not really. The effect is pretty small. Where you might run into it is in high latency links (think satellite) where the MTU (maximum transfer unit) is small, which magnifies the impact of the larger overhead IPv6 requires. That's an edge case. The other area where you'll see impacts is when you're doing 6 to 4 translations in the network path, as that always takes some time. But if you had a pure v6 path to that other v6 host such latencies won't be an issue.

In these days of TCP Offload Engines coming built in to more and more network stacks the impact is even less likely to be noticed. If any. In fact, it may even be faster in those cases.


Why is that larger header not as much of a factor as you think? That's because the designers of v6 took some of the lessons of v4 and built things better. Most importantly for cross-internet communications the address fields are handled much more efficiently in routers than in v4, which improves speed of v6 packets through routers as compared to their v4 cousins.

When it comes to same subnet communications where router tables aren't a concern, each packet requires less raw computation. There is one less checksum to validate (Ethernet checksum, no IP checksum for v6, but TCP/UDP checksum is still required) which saves small amounts of time. And on special networks, the ability to have VERY large packets can further save processing.

0
15

You might see performance improvements with IPv6 if there is a slow router between you and your destination which would have fragmented IPv4 packets along that path. With IPv6, routers will no longer fragment packets for you. (the responsibility has been delegated to end nodes.)

Of course, this is an edge case. There is no reason to believe IPv6 will perform better than IPv4 in the general case.

Also, router vendors may also have optimized their IPv4 data paths more than their IPv6 data paths. So until IPv6 routers have equivalent optimizations, IPv6 may be slower. (For example, some routers may do IPv4 routing in hardware, but IPv6 in software. High-speed performance tests would have to be done to identify this.)

So you have to weigh the chances that there is a router that isn't optimized for IPv6 between you and your destination with the chances that there is a router doing fragmentation of IPv4 packets between you and your destination. ;-)

12

I've been running dual stack IPv6/IPv4 for a while now, and I'm a huge fan.

I can tell you that most of the time the performance and latency are about the same. But on occasion, IPv6 will make your jaw drop.

Consider my ping times to Google:

On IPv4:

rtt min/avg/max/mdev = 35.557/38.225/43.909/3.146 ms

On IPv6:

rtt min/avg/max/mdev = 39.202/40.923/43.105/1.501 ms

Nearly identical. That's what I see most of the time. But every so often, such as when I'm accessing one of my own servers, which is also on IPv6...

On IPv4:

rtt min/avg/max/mdev = 123.510/124.249/125.997/0.909 ms

On IPv6:

rtt min/avg/max/mdev = 87.140/88.600/90.727/1.255 ms

The big difference here, of course, is my ISP's less than optimal IPv4 routing for this route and much better IPv6 routing. (IPv4 goes Boston-Chicago-Denver-Seattle-LA-Phoenix. WTF? IPv6 goes Boston-DC-Atlanta-Dallas-Phoenix.) I suspect there are lots more of these out in the wild.

4
  • Your own server's example is clearly due to routing, and not inherent to IPv6 vs IPv4. Your ISP may even fix the convoluted IPv4 route if you contact them for support.
    – StockB
    May 9, 2016 at 13:20
  • 2
    @StockB Yes, it's clearly due to routing, and I even said so! May 9, 2016 at 13:23
  • 4
    I see that. I'm just clarifying that this is not due to IPv6, and this part answer is therefore inconclusive.
    – StockB
    May 10, 2016 at 13:57
  • 1
    I think it's good, it really shows the YMMV of IPv6 vs v4 in either direction based on your actual setup (and the setup of the powers-that-be!) in connecting to whatever you're connecting to (internet or not)
    – Hashbrown
    May 5, 2020 at 9:30
10

In my experience of comparing traceroute6 vs. traceroute, IPv6 is faster, even through a tunnel broker. This is probably because IPv6 generally requires fewer hops to get from point A to B.

Here's a third-party source backing up my observation:

Accessing sites and content over the Internet is usually faster with IPv6 than with IPv4 because with the new protocol it requires fewer "hops" between network nodes
Here is a little experiment I ran using traceroute/traceroute6, ping/ping6, and my Hurricane (tunnelbroker.net) tunnel, gathering the sites with both IPv6 and IPv6 capabilities from http://test-ipv6.com:
SITE                    IPv4 hops       IPv6 hops       IPv4 ping (min/avg/max/mdev ms) IPv6 ping (min/avg/max/mdev ms)
6connect.com            14              8               67.253/72.499/106.113/11.265    55.031/58.340/74.078/5.426
bind.com                13              10              45.450/53.924/98.121/15.158     60.002/82.812/196.221/41.448
comcast6.net            >13             10              -                               92.043/92.840/94.055/0.680
delong.com              11              7               59.059/66.432/95.884/10.134     52.423/73.042/135.103/30.874
mozilla.org             >11             6               33.178/97.481/536.719/148.432   45.562/47.133/48.390/0.805
test-ipv6.chi.vr.org    9               9               85.383/286.941/527.103/155.351  76.788/169.458/516.502/147.166
test-ipv6.com           10              5               34.021/39.507/70.518/10.384     33.009/41.441/70.052/13.069
test-ipv6.iad.vr.org    15              13              86.739/93.772/120.192/10.195    91.341/93.146/97.153/2.006
test-ipv6.motd.be       >14             8               86.186/401.432/1629.098/502.373 92.437/481.830/727.557/241.649
test-ipv6.sjc.vr.org    12              6               35.443/40.502/70.426/10.056     33.953/41.144/85.444/14.862
tunnelbroker.net        >9              >4              39.504/46.119/73.483/10.290     32.559/63.532/222.700/62.022
So, indeed, IPv6 uses fewer hops and is overall no worse, if not slightly better, performance-wise than IPv4, even though my IPv6 is going through a tunnel.

2
  • Your third-party source does indeed make the claim that IPv6 requires fewer hops between nodes, but it doesn't explain why. Does anyone have any insight on this?
    – StockB
    May 9, 2016 at 13:25
  • 4
    I did some research and found a source for empirical analysis of IPv4 and IPv6 hop counts, which suggests that although hop counts are lesser for IPv6 than for IPv4, this number is increasing over time, suggesting that the relatively low hop counts of IPv6 may be attributed to the fact that their are simply less IPv6 nodes, and considering hop counts without evaluating overall latency may be a meaningless metric.
    – StockB
    May 9, 2016 at 14:28
0

No, there would be no speed gains to speak of, although routing tables are certainly made simpler.

To say that IPv6 "just" makes more room on the net is like saying the Milky Way Galaxy "just" makes more room for dust grains. IPv6 essentially makes it so that it's literally impossible to run out of IP addresses. Even if we were to carpet the entire surface of the earth with CPUs, each with its own IP address, we'd still come nowhere near IP exhaustion. The number of IP addresses in v6 is totally incomprehensible.

12
  • 1
    Can I quote you on that? I'd like to bring it up when every toaster, refrigerator, oven, microwave, pencil sharpener, car, tractor, house, and human being has their own IP and we're looking to expand via IPv8. :P
    – Aeo
    Feb 10, 2011 at 18:54
  • 4
    You sure can. If you wanted to, you could assign 2000 IPv6 addresses to every square meter of the disc of our galaxy. There are 2^128 possible IP addresses in this scheme, or over 3x10^38. This is more than a billion billion times the total number of IPv4 addresses. You could even assign IPs to every single component of every single household item ever made in the entire history of humanity, until the end of humanity itself. Feb 10, 2011 at 20:25
  • 2
    But it is essentially just to add more IP's? ; ) That is a lot more than I imagined v6 adding, and I thought it added quite a bit. Still, the adage is every time you give someone a mile of rope they realize they needed ten, with double the weight capacity. IP's on a molecular level with nano-tech could eat them up, who knows. (Then again I suppose you could nat that to your local organism IP)
    – Jane Panda
    Feb 11, 2011 at 13:50
  • 1
    @ErnieDunbar - before the end of which year?
    – Nas Banov
    Jun 12, 2016 at 1:00
  • 1
    Heh. 2011 was actually the year that we ran out of IPv4 addresses (the last /8 was assigned in January, there was a ceremony and everything). I expected a bigger boom, to be honest. On the other hand, Asia is already moved over to IPv6, and if you have a 4G or LTE phone, then you're also using it that way. Jun 22, 2016 at 17:28
0

IPv6 is not 'faster' than IPv4. If your ISP have a better IPv4 BGP peers than IPv6, IPv4 latency is lower than IPv6. And if your ISP have a better IPv6 BGP peers than IPv4, IPv6 latency is lower than IPv4.

2
  • 3
    Peering and BGP would be a whole different story than the use of IPv4/IPv6.
    – Seth
    Jan 10, 2017 at 14:21
  • 1
    Might be super cool to back that up with some statistics, preferably publicly available.
    – Journeyman Geek
    Jan 10, 2017 at 14:23

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .