[nycbug-talk] Apache, ftp, samba, etc....
bob at redivi.com
Mon Oct 3 03:23:48 EDT 2005
On Oct 2, 2005, at 6:15 PM, Isaac Levy wrote:
> On Oct 2, 2005, at 10:11 AM, Francisco Reyes wrote:
>> I like that idea. Specially for files one does not expect to change.
>> I already have a little script to use mtree to compare directories.
>> How about CPU overhead?
>> I like the concept of a jail, but in the past I always wondered if
>> the extra complexity and CPU overhead were necessary for my needs.
>> I think a current box I am setting up is the first time I think it
>> make sense.
> Re. CPU overhead, it's VERY minimal for jailed systems- by design.
> Jail is not a full-fledged virtual machine, so the resources
> soaked, are simply whatever you end up running in the jail itself-
> jailing is different than Xen or VMware in this respect, as the
> virtualization of memory and etc.. hardware interfaces, are
> comparatively much more taxing- though in the context of the
> applications used, it's all pretty moot... (3k vs. 300k is nothing
> on a machine with a gig of ram ;)
In either case, when was the last time you did anything CPU bound?
Almost every service you're going to run is IO bound by disk or
network (or even RAM). The CPU almost always has plenty of cycles to
The memory usage or increased memory usage might matter, but it's
likely to be quite negligible. Jails are by design basically the
lightest weight method to solve the virtualization problem (since
there is only one kernel for all jails)... Anything that does machine
emulation like Xen or VMWare is going to have to just allocate a big
chunk of memory for each virtual system, so will require a lot more
RAM and memory bandwidth than a jails solution.
More information about the talk