[CentOS-docs] Contributing Wiki article on tmpfs

Wed Nov 4 22:01:28 UTC 2009
Jasper Siepkes <jasper at siepkes.nl>

I think what they mean by that is the following: The Out of Memory
handler will kill processes to free up RAM and swap. However tmpfs'
memory use is not caused by a (killable) process, so the OOM handler
can't free it up.

Anyway when the OOM handler kicks in you have a serious problem anyway,
tmpfs or not ;-)

On Wed, 2009-11-04 at 16:47 -0500, Brian Mathis wrote:
> Looks good.  I appreciate the discussion on the priority of the RAM
> space used and issues to think about as far as swap space size and
> when it starts using disk.
> 
> One thing that concerns me in the tmpfs.txt file is this statement:
>     If you oversize your tmpfs instances the machine will deadlock
>     since the OOM handler will not be able to free that memory.
> I assume that means you shouldn't allocate more to tmpfs than you have
> physical memory, but I've also seen other sites suggesting to use
> size=100% (vmware server tuning).  So far I think 100% would be fine,
> just not 110%.  I'm not sure if you have any further insight into
> that.
> 
> Anyway, thanks for taking the time to address my comments.
> 
> 
> On Wed, Nov 4, 2009 at 4:19 PM, Jasper Siepkes <jasper at siepkes.nl> wrote:
> > Hi,
> >
> > First of all sorry for the long delay and thanks for the constructive
> > feedback.
> >
> > I've made some additions to the Wiki article
> > ( http://wiki.centos.org/TipsAndTricks/TmpOnTmpfs ):
> >
> > -Added the 'Practical details' section to explain how tmpfs, swap and
> > memory relate.
> > -Added a 'Pitfalls' section to warn people of an (IMHO) somewhat
> > misguided piece of advice in the official documentation and maybe some
> > other pitfalls when I think of them.
> >
> > About ramdisks; I will add something later about how tmpfs relates to
> > other ramdisk implementations like cramfs and squashfs. But I haven't
> > done much with the former two so I will need to research those a bit
> > before I can write something down.
> >
> > I'm not the best writer on this continent (Shakespeare beat me by half a
> > point, but I suspect foul play ;-) but I think I've made some steps in
> > the right direction with the article. Please let me know what you think
> > and where I should make some additional improvements.
> >
> > Regards,
> >
> > Jasper
> >
> > On Tue, 2009-10-20 at 14:38 -0400, Brian Mathis wrote:
> >> On Tue, Oct 20, 2009 at 1:46 PM, Jasper Siepkes <jasper at siepkes.nl> wrote:
> >> > Hi,
> >> >
> >> > I've made a first attempt at writing the tmpfs CentOS Tips and tricks item
> >> > and I'm looking for some feedback. Does it need more (or less) background
> >> > info ? Or more (or less) substance perhaps ?
> >> >
> >> > Regards,
> >> > Jasper
> >> >
> >>
> >> I'd like to see more info on things like:
> >> - What is the relationship with swap?
> >> - How/when does swap get used instead of RAM?
> >> - How do I determine the size of RAM to use?
> >> - What impact does using 100% of RAM have on the active memory in the system?
> >> - What happens if all system RAM is full, and tmpfs is also full?
> >>
> >> I think your doc is a good start, but as of now it's got the same info
> >> as all the other tmpfs docs out there, and they don't go into much
> >> detail about the real impact on a system when using tmpfs.
> >>
> >> Tmpfs is billed as a more sophisticated ramdisk, so docs addressing it
> >> should go into the details of what makes it better.  A simple ramdisk
> >> just allocates a chunk of RAM, so why is tmpfs better?
> _______________________________________________
> CentOS-docs mailing list
> CentOS-docs at centos.org
> http://lists.centos.org/mailman/listinfo/centos-docs