Ok Wido, thanks for the clarification, it sounds good.

>From my point of view, if there is a clear relation between IPs and VMs, then 
>as an operator I am happy.

Lucian

--
Sent from the Delta quadrant using Borg technology!

Nux!
www.nux.ro

----- Original Message -----
> From: "Wido den Hollander" <w...@widodh.nl>
> To: "Nux!" <n...@li.nux.ro>, dev@cloudstack.apache.org
> Sent: Thursday, 28 April, 2016 11:41:30
> Subject: Re: IPv6 progress in Basic Networking

>> Op 28 april 2016 om 10:50 schreef Nux! <n...@li.nux.ro>:
>> 
>> 
>> Hi, see in-line
>> 
>> --
>> Sent from the Delta quadrant using Borg technology!
>> 
>> Nux!
>> www.nux.ro
>> 
>> >>Since we know the MAC address and the /64 prefix we can *calculate* the 
>> >>address
>> >>a Instance will take. With that we do not have to store the address in a
>> >>database.
>> > 
>> > Doing bit of research on this makes me believe that this is a preferred 
>> > way of
>> > getting ipv6 address for the instance. Was wondering why we cannot do the 
>> > same
>> > for ipV4, but there I guess we will run into uniqueness issue.
>> 
>> Please feel free to correct me, but SLAAC obtained IPv6 addresses will not
>> always be determinable based on MAC address.
>> Last I looked at this - because we wanted to somehow use IPv6 in SG zone and
>> failed :> - some Linux distros and Windows versions can enable privacy
>> extensions (net.ipv6.conf.all.use_tempaddr).
>> 
> 
> Yes, so a few things are required for the Instances/Templates running. Just 
> like
> a Instance now has to use DHCP for IPv4 it will have to use SLAAC for IPv6.
> 
> All installers right now already support SLAAC. Keep in mind that even if you
> use Privacy Extensions your system will *ALWAYS* generate the SLAAC address
> based on the MAC.
> 
> Take my desktop here for example:
> 
> wido@wido-desktop:~$ ip addr show dev eth0
> 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP
> group default qlen 1000
>    link/ether c0:3f:d5:68:28:08 brd ff:ff:ff:ff:ff:ff
>    inet 192.168.5.70/24 brd 192.168.5.255 scope global eth0
>       valid_lft forever preferred_lft forever
>    inet6 2001:980:XXX:0:f1ac:3224:c00d:47e4/64 scope global temporary dynamic
>       valid_lft 7197sec preferred_lft 7197sec
>    inet6 2001:980:XXX:0:c23f:d5ff:fe68:2808/64 scope global dynamic
>       valid_lft 7197sec preferred_lft 7197sec
>    inet6 fe80::c23f:d5ff:fe68:2808/64 scope link
>       valid_lft forever preferred_lft forever
> wido@wido-desktop:~$
> 
> It generates the temporary address, but next to that it also generates the
> static address.
> 
> For Windows it is easy as well:
> 
> netsh interface ipv6 set privacy state=disabled store=persistent
> netsh interface ipv6 set global randomizeidentifiers=disabled store=persistent
> 
>> So for the scenario to work, we need to make sure privacy extensions are
>> switched off - but who knows for how long this will be doable, not to mention
>> some ISO, PCI etc policies might require it on.
>> 
> 
> As CloudStack we do not have 100% control of the Instances. Same story with 
> IPv4
> right now. Our docs should just say how to configure a Instance.
> 
> But with IPv6 you should also let go of the single address. My proposal is 
> also
> to enable Prefix Delegation:
> https://cwiki.apache.org/confluence/display/CLOUDSTACK/IPv6+in+Basic+Networking
> 
> Next to the /128 (Single Address) a Instance can get a /60 where it can pick 
> as
> many random addresses from as it wants.
> 
>> Some random reading from the interwebz, he touches on the subject:
>> https://major.io/2016/04/17/enable-ipv6-privacy-networkmanager/
>> 
>> 
>> IMHO - without knowing all the ugly details behind DHCPv6 and so on - I think
>> going for DHCP might be the better option in the long run.
>> 
> 
> DHCPv6 is ugly and not properly supported. If we go down that road we suddenly
> have to:
> - Store a IPv6 address in a database
> - Provision a DHCPv6 server
> 
> That's additional steps which can fail and make stuff more complex. KISS is 
> want
> you want.
> 
> SLAAC is supported in Linux, *BSD and Windows with EUI-64 based addresses, 
> that
> is doable.
> 
> Again, with IPv4 we currently also require various settings inside the 
> Instance.
> That will not be different with IPv6.
> 
> Wido
> 
> > Lucian

Reply via email to