All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Acceptable CPU steal?
I have a few different VPS's with different providers across Europe with similar specs using KVM virtualisation. Over the past week I noticed the load average increase significantly, and then looking at the CPU 'steal' metric, it's gone from previous weeks of only being a few percent to now in the 15-17% range and staying there. Because of this the load average is also out of the preferred range as well.
I had this problem on another host recently, and was moved to another physical host - CPU steal went to 0 immediately and hasn't been a problem since. The new machine was pretty much empty.
My program is time-sensitive so it's key to not miss CPU cycles, I'm always trying to keep my load average less than 0.7 (normalised) - so just wondering, what would you consider an acceptable level of CPU steal for a VPS? Is there any kind of virtualisation technology where CPU steal isn't possible at all?
Thanks
Comments
Have you got dedicated cores on your VM or shared?
You hit 10%+ open a Ticket and request them to check on the Node or/and migration.
You expect them to monitor the nodes, I am sure most of them do however due to the pricing might be only if you complain you get it fixed.
Expectation should be 0% steal unless your own vps is restricted.
Pretty sure they would be shared cores.
0,.. %
steal average <5.0% if Black Friday Cheap VPS.
<1% if lowend VPS.
I think I need to write in another ticket. It's consistently over 10% (and sometimes as much as 20%) over the past week so this is really starting to affect the performance of my program.
Probably a time to yell at contabo for me
you get what you pay
still acceptable,, right ?
Yes.
No. I know that contabo is cheap af, iops limited, support on OVH levels (or even worse tbh, OVH straight out ignore you, contabo play dumb), but you expect them to monitor things like that - one day... sure. Two days - over weekend? Maybe, it's cheap, weekend salary is expensive.
This is avg for last 7 days so no, not accaptable.
I don't feel so bad now
maximize RAM usage to reduce CPU
such as varnish and other cache
longer execution time in php and web server
maybe
contabo still ok
Well, in the end, it depends what you do and how much it is.
Anything that gets directly impacted by it, Webservers not so much, but for example in my use case, forwarding network traffic, it does impact it, measurable.
I can see on the local smokeping, that the machine despite 1% cpu load is having a hard time with the cpu steal.
Guess who got migrated.
Nice. What's your avg steal at now? This is Contabo isn't it?
But this is for like 1.5 hour of running.
Yes, contabo, cheap thing.
and world wide stable network ?
That is my contabo machine xD 35+% cpu steal
so .. contabo expensive price... right ?
BF-2020 AMS $6.4/yr
Smokeping before and after migration, Gcorelabs Moscow
The provider I was having trouble did address the issue - but only for 24 hours I or so. CPU steal dropped to about 2% for the past day, now it's back to about 12% - they say nothing can be done about it, and that I can try another server and see if I'm lucky enough to find a less busy one. I think this is grounds to move host...? I'm fine with a couple percent CPU steal but this 12% business is just too much for my application. I have a small VPS with Linode and the CPU steal is at 0 nearly all the time. Anyone know any VPS hosts in Germany that don't oversell their resources and have competitive pricing?
How can you measure cpu steel on a live DA VPS please?
If Netherland works, come check out EntryBytes.com boss
It should be less than 5% all the time, ideally less than 2%. Very high CPU steal usually means the provider is overloading that node and they don't have enough CPU power to handle the number of guests running on it.
It appears in
top
as thest
value, and invmstat
as thest
header.It's the 8th field in
/proc/stat
for thecpu
lines - this is documented in the proc(5) man page: https://man7.org/linux/man-pages/man5/proc.5.html. All tools that read it get the data from /proc/stat.Not bad for @contabo_m after they moved me to other node. Still seeing odd 10% or something from time to time, but this is how shared works and I can live with that.
Same things running almost same conditions.
@HBAndrei always forget to tag you - is there any chance for
steal
in Server Monitoring Agent on the cpu graphs? I know VPS's are probably not the most lucrative clients, but still could be helpful from time to time :-)Thats still pretty high, I'm on a underloaded node on contabo probably:
Plus someone accidentally enabled inside virt for me lol
No, your CPU usage is almost zero [%usr 0.31] (you are and idler, perfect client!) and there is just not enough CPU demand to even get into 'steal' part - you demand almost 0 CPU, host node is not having problem with allocating that to you. More steal will show up with more CPU usage :-)
Depends on how many VPSes the host cram onto a single server I only get >3% with one particular provider, all the others are usually less than 1% apart from spikes, and this includes small NAT VPSes too
On a HostHatch VPS in particular, I'm using ~15-20% CPU constantly and the steal% is almost always below 0.3%. On a BuyVM VPS, I'm also using ~20% CPU constantly and the steal% is usually 0.7 to 0.9% (I was genuinely surprised that HostHatch's older E5-2680 v2 servers have less CPU steal than BuyVM's new Ryzen 3900X servers)