UbuntuIRC / 2012 /03 /07 /#ubuntu-cloud.txt
niansa
Initial commit
4aa5fce
=== dendro-afk is now known as dendrobates
=== Hussain is now known as Guest28163
=== marrusl_ is now known as marrusl
=== dendrobates is now known as dendro-afk
=== dendro-afk is now known as dendrobates
=== Hussain is now known as Guest21263
=== dendrobates is now known as dendro-afk
=== dendro-afk is now known as dendrobates
[21:23] <thomi> is it still the case in precise that I can't use juju to deploy to a rackspace cloud instance?
[23:09] <SpamapS> thomi: There's no RS cloud provider been written, no.
[23:09] <thomi> SpamapS: ok, thanks.
[23:11] <nOStahl> hi guys
[23:11] <nOStahl> i got my hard drives in today
[23:12] <nOStahl> two caviar black 1.5 terabyte setup as raid1 in a hp dc5750 tower
[23:12] <nOStahl> I am wondering if I should use openstack or eucalyptus to setup a cloud?
[23:13] <SpamapS> nOStahl: on a single server?
[23:13] <nOStahl> i have dell optiplex 760 two if them for the nodes
[23:13] <SpamapS> nOStahl: I would go with OpenStack. It seems to have the brighter future and is receiving more attention on Ubuntu.
[23:14] <SpamapS> nOStahl: neither of them are very easy to deploy though
[23:14] <nOStahl> I want to be able to have a production instance for web server , email, calendar , contacts servers
[23:15] <nOStahl> and be able to launch an instance for testing stuff etc.
[23:15] <nOStahl> all small stuff max of 50 emails low usage etc
[23:15] <SpamapS> nOStahl: if you just want basic virtualization.. both of them are pretty much overkill.
[23:15] <nOStahl> so openstack eh?
[23:16] <SpamapS> nOStahl: libvirt will probably do most of what you want for what you just described.
[23:16] <SpamapS> nOStahl: clouds are more about being able to scale much much bigger. How many physical CPU's and GB of RAM do you want to support long term?
[23:17] <nOStahl> aye thats what I was thinking
[23:17] <nOStahl> can add nodes as needed
[23:17] <nOStahl> and for redundancy
[23:17] <nOStahl> thats a question I was having btw
[23:18] <nOStahl> if I launch an instance and its running my webserver
[23:18] <nOStahl> and i have two nodes and one dies
[23:18] <SpamapS> nOStahl: openstack has no real notion of redundancy from what I've seen. Eucalyptus just now added HA in their 3.0 release.
[23:18] <nOStahl> the instance still runs?
[23:18] <SpamapS> nOStahl: hahaha
[23:18] <SpamapS> hahahahhaha
[23:18] <SpamapS> sorry I'm not laughing at you ;)
[23:18] <SpamapS> nOStahl: that sounds pretty magical though, doesn't it?
[23:18] <nOStahl> yes it does
[23:18] <SpamapS> magic is bad mmkay :)
[23:18] <nOStahl> heh
[23:19] <nOStahl> is what I'm talking ha?
[23:19] <SpamapS> nOStahl: there is no magic bullet for node HA. You're going to lose something if you make things available. See google searches for "CAP Theorem"
[23:21] <nOStahl> so run me through how it works
[23:21] <nOStahl> i launch an instance and it pushes the image over the network to one node? or many nodes?
[23:21] <SpamapS> nOStahl: one
[23:22] <nOStahl> ah I was thinking that it kind of mirrored it some how
[23:22] <SpamapS> nOStahl: can you imagine how slow it would be if it mirrored all RAM access?
[23:23] <SpamapS> nOStahl: there are "live migration" capabilities, where you can move nodes from one machine to another for expected downtime.
[23:24] <SpamapS> nOStahl: and if you are using nova-volume with iSCSI, your disk will be stored somewhere else, so upon dying, nova can re-start your instance somewhere else.
[23:24] <nOStahl> that must be what I was thinking of
[23:24] <SpamapS> nOStahl: for the most part, you need to think about failover and recovery at the OS level, not the cloud provider level.
[23:25] <SpamapS> nOStahl: DRBD for your data works
[23:28] <nOStahl> getting more complicated lol
[23:33] <SpamapS> nOStahl: scale out is complicated. :-P
[23:38] <nOStahl> so ubuntu 11.10 then?