On Wed, 2012-03-07 at 10:12 +1300, Jim Cheetham wrote:
> On Wed, Mar 7, 2012 at 2:26 AM, Kent Fredric <[email protected]> wrote:
> > On 6 March 2012 12:36, Adrian Mageanu <[email protected]> wrote:
> >> Yes, disk I/O mostly due to furious swap when available RAM is exceeded,
> >> and then there is the 100% of every available CPU, that's when keyboard
> >> and mouse get locked and the fans screm mad.
> 
> Swap is overrated. These days we have sufficient RAM to keep the OS
> running properly, and when machines descend into swap the machine
> generally dies because we overcommit and have slow disks. Switch off
> swap, and let the OOM killer do its job - on a well-run server a
> critical service should be restarted by init anyway ...
> 
> $ free
>              total       used       free     shared    buffers     cached
> Mem:       8149024    7968776     180248          0     192868    4736620
> -/+ buffers/cache:    3039288    5109736
> Swap:            0          0          0
> 
> More of this philosophy is touched on during Anthony Town's recent
> LCA2012 presentation :
> http://www.youtube.com/watch?v=p_u6BDFkybE
> 
> -jim

You're suggesting that it's OK to let the system kill off critical
processes?? That's a bit of a DJB approach imo...

What if that process removes all connectivity with said server? And said
server is hosted in Guatemala? And the hosting provider has no staff to
support you until Monday morning? 

Yes, this has happened to me ( Well, it was Canada, but I like typing
Guatemala... ). Suffice to say I no longer use them ( I've listened to
recommendations, and now almost exclusively use linode for international
sites ) but that's by the by.

It is very annoying and difficult to plan for a swapless scenario -
there's (at least) one local VPS provider that does this, and setting
the server up, creating alarms all over the place just to be sure you
can cope is a real PITA. You can easily end up with services that say
they're up and running when they're stuffed, so just using inittab is no
good - you end up using ping/pong type programs just to check they
really are up. More overhead, more testing... generally more grief.

My big gripe is this. I'm typing this on my new workstation. The memory
cost $50 for 8GB. Why doesn't everyone just stuff servers as full as
possible, with the intention of never using swap, but having it there
just in case. This is extremely useful, for example when commissioning a
new server, and you need to tune your guesstimate config to what's
really required.

My $0.02,

Steve



-- 
Steve Holdoway BSc(Hons) MNZCS <[email protected]>
http://www.greengecko.co.nz
MSN: [email protected]
Skype: sholdowa

Attachment: smime.p7s
Description: S/MIME cryptographic signature

_______________________________________________
Linux-users mailing list
[email protected]
http://lists.canterbury.ac.nz/mailman/listinfo/linux-users

Reply via email to