New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
These screens dont show performance nor CPU steal AT ALL. These screens show metrics of CURRENT USAGE of provisioned resources, nothing else and nothing more.
So OP's VM is idling and all 4 CPUs should spike in random intervals
Wondering why my idlers don't do that
because host scheduler doesnt give a fuck about virtualisation layer and places tasks to different cores that are most suitable to new task (Core2core latency, CCX latency, thermal limit etc).
Providers may pin your cores to one NUMA, but thats all.
Dedicated core = you can use it 24/7 without problems
Dedicated network port = you can use it 24/7 without problems.
If you need 100% pefromance you choose dedicated server and your own network cables. Dedicated is NOT guaranteed. It means that they allow you to use this resource and thats all. No guarantee of X performance.
@AXYZE
Is there any implication that "dedicated" means that no more than the amount of host cores are sold to customers as dedicated VM cores? Or supposedly it just means that 100% CPU usage is permitted, regardless of the amount of VM cores versus host cores?
????????????????
You couldnt mix it up more. Where did you get such info?
If he has CPU spike it means that his own OS/app requires CPU at that time.
Just look up task manager on your PC, its same thing.
Problem with CPU steal is that instead of fast spikes to calculate what is needed you need to wait for CPU resources so task will take longer to do and you will notice longer load that takes less % per second.
I just lost track of everything tbh. Sorry if I misunderstood anything. Just want to recall that OP said that he's not running anything on the VM and it's idling.
My experience is that I've never seen a VM spike that much being idling.
It just means that 100% usage is permitted unless they wrote that CPU is guaranteed, only then you can be sure that your VDS is pinned to some cores. Never saw provider doing that as its not cost effective at all and its better to just sell whole dedi.
As all the nice & experienced people in here suggested:
It is normal for such servers to work like this after a few weeks. Just don't expect the performance or benchmarking results to be the same, nothing is guranteed.
You can see my Hosthatch VPS top results for cpu stealing. its a 8 core vps (4 dedicated and 4 shared), 39$/month non promotional VPS.
It becomes 40–48+ in the night, and I face performance issues. I have opened a ticket and have been waiting for a solution for the last 40 hours. I also purchased this VPS 1 month ago. After purchasing it, it was rock solid; there was no stealing at all. My application performs so fast that I feel like I'm on a dedicated server. Anyway can you please explain weather it's normal or not?
Yesterday top result:
CPU spike has nothing to do with CPU steal.
It means his OWN systems demands CPU and CPU was available for this spike to happen.
If he has problem with spikes he should just look up what app in his own OS is causing it.
For example something he installed can call unexisting services over and over causing these spikes - problem that my friend got some time ago, because some old app was trying to run php7.1fpm over and over 50 times a second.
No, its not normal. This is real CPU steal that causes severe system latency.
Depending on app it may be noticeable (for example database) or not at all (for example serving static files with nginx). I see that you run MySQL so queries with such high steal will be dozen times slower (more than steal suggests as youre dealing with latency issues here), but it can be still fast enough (for example 0.1ms vs 1ms) or your most common queries are cached either by plugins, web server or MySQL in-memory cache.
I would suggest to contact HostHatch.
CPU steal above 20 is unacceptable in anything more than $2/mo VPS.
Just tried the command you suggested previously:
Not sure if its considered bad or good.
Exactly
But also, before causing the provider the headache, think about if you actually benefit from the resources
So many people are constantly complaining but in reality run NOTHING except for benchmarks
Pretty much you got noisy neighbour, I dont know how they manage pool with dedicated + shared core, but that looks horrible even my contabo not have steal like that
Hope you got resolve pretty soon.
Idling? If so then try to run some benchmark like Geekbench 6, and see the CPU Steal, surely it will be higher. CPU Steal is not good if it's above 10 (constant average). Especially if you are running intensive task like database queries, it will be slower for sure, comparing with "almost" zero Steal VPS.
Hosthatch
dedicatedcore means you can use 100% of it, not that it'sdedicated(exclusive) to your VM (it's still shared with other people). Kinda missleading, but this is how it works (at least for offer on this forum, I've asked ;') )Steal 30% is no good.
5% steal on idler is no good either. Steal happens when your VM wants to do something, asks for CPU, got denied... So the more you run, the more you ask, the more steal you will get.
Because all those servers are shared (one way or another) - CPU/RAM/disk etc. shit like that will always happen - however it's providers job to monitor those things and react to those - limit/kick an abusive neighbour on your node, investigate things etc.
If someone decided to run some very CPU intensive program for 5 minutes and it generate some steal on neighbours you just go with it - it will happen from time to time on shared. If that goes for 40+ hours then
If it happens for 40+ hours and support thinks everything is okey then run ASAP.
btw. Everyone oversold VPS, that is the only way to get profit for those - you calculate some 'avg' uses, you put the number of people on node that they won't overload the machine (like 99.95% of users don't use 100% CPU 24/7). Some oversold more, some oversold less, some can just have a machine "abused" (or overloaded) because half of people on that node found out this new cryptoshit and they just run 100% CPU out of blue... after months of 2.5% avg. This will happen, always, anywhere on shared machines - it's the provider reaction you judge the quality of host.
also just open a ticket with provider - maybe it's not an abusive user and it's something they know about and they just didn't wanted to message all users with "You can have degraded performance for next 48 hours as one of the disks dropped out of array and rebuild process takes some time" etc. - most of people (2.5% CPU avg fe.) won't notice.
thanks for you detailed explanation. My entire database is cached on RAM and that's the reason my application is still running. But when st goes above 40 i start facing problem. specially grabing record from the db as it is a dynamic app it takes time to create the cache after receiving new data.
Don't get me wrong, but for benchmarking, I will never purchase a $39/month plan, just 5$/month will do the work. I need power (not always), which is why I purchased a bigger package. I may or may not consume the resources, but they should be available for sudden visitors/data spike. Normally, I do not require this many resources. as I have optimized my database to mainly work from the RAM. But still, I need power to get piece of mind that my app will not crush if there is any sudden load. i have a busy application there
For example my server mysql stats (as per phpmyadmin):
Uptime: 1 days, 15 hours, 7 minutes and 28 seconds
Questions:
ø per hour: 29,863
ø per minute: 498
ø per second: 8
last 24 hour records (Cloudflare):
Total Requests
103.98k
Percent Cached
62.16%
After purchasing hosthatch my server load was less than 0.60 all time for first few weeks. now it's more than 1.5 (At night 2.5). Still it's okay but this server will definitely fail if i get sudden visitor spike. that's what i don't want and thats the reason i have purchased more resources than i use. i select HH because they have offered 4 dedicated core and 48GB ram. But i made a mistake. i will increase the budget and find any better provider. thanks anyway.
I would heavily suggest GreenCloud VDS in similar budget
https://greencloudvps.com/epyc-nvme-kvm-vds.php
I have 3x VPS from GreenCloud and all of them have steal less than 2%.
Amazing performance, VERY fast support, good price.
Thanks for the suggestion. actually, I have talked with them, they use cdn77 and it has latency problem. CDN77 route traffic from Singapore to Bangladesh via India. Most of the time packet loss occurs at india location. This was another reason for selecting HH.
For example rack central
from cdn77:
OVH, Contabo (use HE i guess), Equinix, Rack central and everyone send the traffic directly to bangladesh except cdn77.
As i said, Speedypage use Datacamp aka CDN77 as their upstream in SG. Or maybe try Kuroit and Webhorizon. They are using RacksCentral, and blend Datacamp (CDN77) - Telstra as their upstreams.
https://sg.lg.kuroit.com
https://lg-sg-sin.webhorizon.net
Yes, you're right. But Speedypage has very good latency and a direct connection to Bangladesh (even if they use datacamp). While @greencoludvps is routing via India using airtel connection. if they route like speedypage i will take their vds plan.
Ah sorry, looks like miscommunication. Speedypage SG use Datapacket network.
Probably this. OP could be on a new node and not fully populated so turbo is still in effect.
I always like to judge my servers by very few metrics:
Is it doing what I want it to do, in the time that I want it to do it? If it isn't, is that inherent in what I'm making it do, or the fault of the server to process it in the time I want it to, given that the time I want it to is a reasonable expectation for the job?
Nothing else matters.
BUT THE TOP COMMAND JARLAND
HostHatch is generally good, but I've also noticed slower performance and stolen CPU peaking around 40% on one of my VPSes in Los Angeles in the past month:
Zooming in to the past month:
(the gaps were unexpected node reboots which I asked them about, and they couldn't explain other than "There was a sudden reboot on the host node but this has been resolved already and should not happen again in the future", only for it to happen again ~2 weeks later)
I did open a ticket too.
I reached out to HostHatch and they said there was another VPS on the same server as mine that was abusive / using way too much CPU. I think they throttled that VPS. It's better now - steal is mostly gone.
Sorry just seeing this thread.
We had a backlog after dealing with the New York fire last month, which led to some abusive VMs not being caught on time.
In any case: you should never see CPU steal on a HostHatch VM. If you do, please open a ticket, and we will fix it for you and give you a reason for why it happened in the first place.
We take pride in the fact that we provide consistent performance on our AMD Compute servers (unlike some of the German providers who market "dedicated CPU" and then you end up competing with all your neighbors on the node for that CPU)