[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On Thu, 7 Apr 2011, Simon Waters wrote:
On 07/04/11 15:36, Gibbs wrote:I don't know a great deal about memory usage but I'm struggling to understand this. On my VPS memory usage is typically between 65-95%. This is fine as its there to be used but I do get concerned when it reaches the 90%+ region for periods. Another thing I don't really understand is that none of it is buffered or cached. Here's the output of "free -m": total used free shared buffers cached Mem: 960 807 152 0 0 0 -/+ buffers/cache: 807 152 Swap: 0 0 0 Can anyone explain if this high usage is OK? I have around 20 websites on this Ubuntu VPS including mail. Amavis, Clam, Apache and MySQL are taking the majority.VPS implementations vary, I assume your VPS is KVM based.
I'm fairly sure that Parallels/Virtuozzo is the one that Gibbs's hosting company is using... I think it's a containery type of thing similar to OpenVZ and LXC with one kernel and lots of containers using the same kernel image.. Actually, OpenVZ seems connected with them too..
I have a diddy KVM virtual host which we use as a DNS server. It shows similar characteristics.
It does surprise me, but I guess that's a "feature" :)
I believe swap for KVM has only recently been sanely implemented, and it may not make commercial sense (they want to sell you as much memory as you might need, and they don't want to create a disk I/O issue for themselves).
Reducing disk IO is the main reason - kills performance of other VM's in the same host...
Afraid the answer to is the high memory use an issue is "how much do you need?". If it ever starts killing processes because it is out of RAM, then you don't have enough (or some process suddenly expanded for some other reason).
The dreaded Oom oops ...
Virtual servers lacking swap support is one of the drawbacks of virtual
servers (well Gordon and his telephones might not care), but for website
and database stuff paging out rarely used pages is nearly always a
performance gain, and a soft landing with paging out is often to be
preferred in low memory situations compared to killing off the next
process that asks for more RAM (always seems to be cron or mysql in my
experience, and losing cron can be painful especially if you use it for
back-ups).
My KVM VPS shows:
# free -m
total used free shared buffers cached
Mem: 200 113 86 0 0 0
-/+ buffers/cache: 113 86
Swap: 0 0 0
I'm using LXC for my hosted VoIP boxes:
$ free -m
total used free shared buffers cached
Mem: 3515 2807 708 0 754 1656
-/+ buffers/cache: 397 3118
Swap: 0 0 0
And I do care about swap - it's the absolute last thing I want for the
VoIP boxes - fortunately they're very well behaved and don't grow at all.
For my general purpose LAMPy ones, (which also use LXC) they do have a bit of swap, but not much - basically it's there to take the pressure of something that does need a bit (e.g. cron backup jobs!) more rather than being intended as cheap memory...
Newer LXCs can limit container memory and swap size. Gordon -- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq