On Tue, 20 Jul 2004, Walsh, Bryan wrote:
Even if you use radvd, your server will always get the same IPv6 address. You shouldn't need to use static addresses. You could also use static addresses on the servers if you like while also running radvd for any hosts that might connect to that network.
Asuming you're always advertising the same prefix from your router, the last 64 bits of the servers IPv6 address will be figured out via EUI-64 (which is basically the MAC address of the server with a bit flip at the beginning and ff:fe in the middle).
Eg, this: 00:20:48:5e:2f:29
becomes this: 0220:48ff:fe5e:2f29
So, you can put a static DNS entry in for the server since it will always have the same IPv6 address so long as you keep the same NIC (mac address) in the server and your router always advertises the same prefix on that network.
But completely agree with Jeroen. You can use autoconfiguration, but for the safe side I would recommend using static address. If your machine fails or NIC fails you can easily migrate service to another machine without touching the DNS.
Regarding your second question, I've seen people just use /64's for everything, even point to point links. I think this is wasteful, but thats probably a personal opinion. There is nothing keeping you from configuring a /127 if you like. I do this all the time testing ptop links.
I would not recommend /127 between routers. See more for discussion: http://www.netcore.fi/pekkas/ietf/old-versions/draft-savola-ipv6-127-prefixlen-04.txt
Hope that helps. Take care. Bryan
-----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of [EMAIL PROTECTED] Sent: Tuesday, July 20, 2004 10:54 AM To: [EMAIL PROTECTED] Subject: IPv6 Best practice
Hello all, I have been "playing" with ipv6 for a while now (mostly on Linux and osX) and I have started to turn my thoughts to networking and servers. The easy one I guess is servers. Presumably a static ipaddress is best to use because of DNS etc. If a static address is allocated, radvd will not be required because there is no ipv4 DHCP type requirement. Is this a correct assumption?
Second, networks. On an ipv4 based ip network, it is usual on wan links (unless they are unnumbered serial lines) to use a .252 or /30 mask with 4 addresses in the subnet (net, ip1, ip2, broadcast). Is this wise to implement in ipv6? eg use a /126 mask to allow four valid ipv6 addresses. In that case, if I get a /48, I would need to use the first allowed block (/49 mask?) carved up into much smaller chunks, ultimately down to the /126's for wan lines.
Given a working ipv4 network where each remote site has a /24 ipv4 allocation (and is more than enough given the number of pc's there), would it be sensible to use a /120 for each site or perhaps be profligate(!) and use /118 to allow for all the ipv6 toasters we are likely to be able to buy next year?
Any thoughts on this would be welcome, there seems to be quite a lot of tech info about, but less on the planning rather than implementation side.
Regards, Andrew
-------------------------------------------------------------- -------------- There are only 10 types of people in the world:- Those who understand binary & those who don't.
--------------------------------------------------------------------- The IPv6 Users Mailing List Unsubscribe by sending "unsubscribe users" to [EMAIL PROTECTED]
--------------------------------------------------------------------- The IPv6 Users Mailing List Unsubscribe by sending "unsubscribe users" to [EMAIL PROTECTED]
--------------------------------------------------------------------- The IPv6 Users Mailing List Unsubscribe by sending "unsubscribe users" to [EMAIL PROTECTED]
