[Lxc-users] Many containers and too many open files

[Daniel Lezcano]

On 02/25/2011 11:19 PM, Brian K. White wrote:
> On 2/25/2011 2:06 PM, Geordy Korte wrote:
>> Maybee a really stupid question... but why would you want to run that many containers?
> I won't say it's stupid but I'll say it's meaningless. Don't take it as
> an insult I'm just explaining it's the wrong way to think about it.
>
> It's like asking why do you need 10 computers or 3 cars or 17 pairs of
> pants or an infinite number of integers in math? What would anyone do
> with those really big numbers higher than say, 1 billion?
>
> The answer is really, since a container is essentially like any other
> computer, a general purpose tool, you don't have any specific number of
> them that you need. Anyone might need any number of them for any reason.
> And you generally need more and more of them forever for all manner of
> different uses.
>
> So what you want to know is simply what are the limits? Most people
> probably don't have any specific magic number that must be met, we just
> need to know where are the various limits, whether that turns out to be
> 20 containers per host or a million.
>
> Also there is the fact that, many times the indisputably valuable "real
> world" uses for things only come about _after_ they have been made
> possible. People thought the invention of the transistor was a pointless
> curiosity at first.
>
> If 1000 (or more) containers is possible, then before long I guarantee
> you someone will invent some non-trivial use for that capability.
>
> And that's not counting the obvious ones we can already think of right
> off the top of our heads like:
> - A service provider being able to run all their customers web sites, or
> as in my case their core back office industry-specific custom software,
> on a single or few machines instead of 1500 machines.
> - Running stress tests like simulating 1500 users or network nodes
> hitting a single service or interacting with each other mesh-like, by
> having 1500 vm's actually do it, instead of the old way of either being
> rich and having a bunch of real machines, or "deploy and hope".
> - Developing and testing tools that manage lots of servers, without
> having to actually have a lot of real servers sitting there idle just
> for that. When I make some new script to do something like run a command
> or modify a config file on all of my production servers at once, I
> _really_ would like to test that on a 1500 vm's instead of the real
> production boxes.

Brian answer's has the best score :)