1 |
On Monday 27 June 2005 06:23 pm, Josh Hunholz wrote: |
2 |
> Wilkins, Vern wrote: |
3 |
> >What would the web interface do, or more importantly, what would it do |
4 |
> > that webmin doesn't do already? I'm not a programmer, but it would seem |
5 |
> > that writing a module for webmin to handle something like portage |
6 |
> > administration would be much easier, and probably more quickly result in |
7 |
> > a more complete administration interface for a gentoo machine. I'm not |
8 |
> > doubting that one could develop a better web interface or tool than |
9 |
> > webmin, for Gentoo, just curious. |
10 |
> |
11 |
> The idea behind this is far more than Webmin does. One of the coolest |
12 |
> features looks to be a way to keep multiple Gentoo boxes up to date with |
13 |
> security fixes (so it will tell you what boxes have what that need |
14 |
> updating, etc.) and also more of a large-scale administration of |
15 |
> multiple servers from one interface (where Webmin does one server). |
16 |
> |
17 |
> --Josh Hunholz |
18 |
> |
19 |
|
20 |
Or, an expansion on the whole idea may come as something like this: |
21 |
A cross-platform distribution (based on Gentoo ;) ) with the ability to |
22 |
cluster (however it be necessary). Comes with tools (written in python?) for |
23 |
image creation, whereas images are "stacks" of packages prebuilt/compiled by |
24 |
the administrator. You might specify the stack or meta-ebuild as a set of |
25 |
compiled packages. Once an image is created it functions as an enterprise |
26 |
component. The image then is run under a virtual machine that supports |
27 |
running on a cluster. There are sets of tools to manage the virtual |
28 |
machines, and allow for primary and backup virtual services to be dynamically |
29 |
loaded and configured, or shutdown when not needed. |
30 |
|
31 |
This creates a dynamic network where you would design it with statefulness in |
32 |
mind. When I say statefulness, I am referring to how state changes in a |
33 |
network. One minute things are fine, the next, there are floods of traffic |
34 |
pumping through your email server and ldap is being floored because you |
35 |
didn't have time to replicate it into a secondary. With that in mind, we |
36 |
specify something as responding within a time limit as maintaining a state. |
37 |
The moment it breaks that limit, action is taken to recover the state. |
38 |
|
39 |
Network operation can be dynamically altered by setting up (virtual) bridges |
40 |
between the "enterprise components" and dynamically routing traffic between |
41 |
each of the components. Each enterprise component would sit as a group of |
42 |
almost duplicate virtual machines dedicated to managing one task. When load |
43 |
goes up, more virtual machines are powered up. When load goes down, some are |
44 |
shutdown. We could even go as far as incorporating AI to preload these to |
45 |
match trends. |
46 |
|
47 |
Some advantages may be: |
48 |
* Redundancy - something goes down, the service can be replicated as more |
49 |
virtual machines are powered on. |
50 |
* Load balance all services as one - Each service is balanced (application |
51 |
specific, or network traffic) across multiple virtual hosts, but sits on top |
52 |
of the cluster. The cluster distributes the load of the virtual machines. |
53 |
* Hardware failure - Hardware fails, nothing goes down? I like that :) |
54 |
* Statefulness - all that needs storage is the states of the machines, and the |
55 |
data that circulates, I'm thinking storage area network... (i.e. providing |
56 |
the state, and a data source, the cluster builds itself to maintain the |
57 |
state, then provides services to circulate the data) |
58 |
* Cost - finally a use for all that old equipment we have laying around |
59 |
|
60 |
This may not be the next step, but the next step might lead to this. |
61 |
|
62 |
Robert Larson |
63 |
-- |
64 |
gentoo-server@g.o mailing list |