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.
Question on Contabo CPU usage limitations
AleksaDjordjic
Member
I've got a quick question regarding Contabo.
They obviously oversell the CPU a lot, what I'm interested in is do they limit you somehow if you use the CPU too much?
I know some providers which limited me if I was constantly using 50% of the CPU, or bursted for a prolonged period of time, and I get why, noisy neighbor, bare metal exists for a reason.
But does Contabo do this?
And to what extent?
Any information on what constant % usage can I have, and for how long can I burst to 100%, has anyone had experience with Contabo limiting or terminating their service for being a noisy neighbor?
Comments
They VPS line is heavily oversell , and i don't think they put any CPU limitation.
rule of thumb is 33% constant use with 100% limited bursts
Sounds fair enough, I'd guess this would especially apply to "cloud" providers like Vultr, DO, Hetzner, Linode, etc?
Saw one of their blogs which mentioned mining on their VPS instances, they said its prohibited because it would make you a noisy neighbor, I'd guess they must have some way of seeing if someone is using 100% 24/7... r-right...?
I have maxed out the CPU on Contabo for 20-24 hours straight while rendering a map. I’m not sure if they limit your CPU after maxing it out for longer, but they might limit it silently.
That's BS. I ran multiple benchmarks (which do go to the limits) and over a thousand times ...
and had no problem at all. On a normal VPS. The only 'but' is that my benchmark intentionally makes (very short) pauses between cycles (because it's designed to be a good neighbour).
The point many fail to get is that a VPS by definition is "oversold" that is, that usually there is more than 1 VPS per HWT. If one really needs a full vCore one should buy a VDS.
Yep, that sounds reasonable as a rough guideline.
To be fair, what probably most, or at least me see as "oversold" is overselling more than the usual provider would. Contabo has the highest hypervisor steal that I have noticed, I have 2 clients on Contabo, one in USA and one in Germany, the USA one is actually quite fine, but the Germany one, boy oh boy, I have seen steal reach ~40% and stay there for a minute, meanwhile average was anywhere from 3 to 10%.
Then again, prices are amazing, even when considering that, so can't really complain much.
When I tried out their Windows VPS last year I did run a benchmark test using 100% CPU/RAM for over a week and noticed within the first 24 hours they did limit the CPU reducing it's overall performance by 50%.
Another thing I noticed about their RAM was they had no ballooning driver installed which is required for Proxmox (they advertise they use proxmox???)
Interesting, do you maybe know in details how it was limited. Did the benchmark just see a dip in performance after 24 hours, and was it sustained, are you sure it might not be something like this where the node was just filled up too much and hypervisor steal was just too high: https://www.vpsbenchmarks.com/trials/834/endurance_runs
About memory, can you link me to where they advertise using Proxmox, never heard that before.
Just did a bunch of quick tests, some hardcore (behaving like a pig) and some normal ("good neighbour" mode). The difference seems to be +-10% to 15%, but actually is to do with normal phenomena related to heavily multitasking processors and VMs.
Counter-check: Running the same benchmarks with the same parameters on my local Ryzen I see pretty much the same pattern but somewhat less pronounced. But then I'm not running dozens of VMs on my box with who knows what workloads.
I used contabo amd cpu at for days(Video encoding) didnt get any warning. But that doesnt mean you wont get either. Its shared anyway.
24 hours later (even after rebooting and reinstalling OS)
Can't find that info about memory just can recall reading it somewhere might be wrong.
Can you run something like
vmstat
and send the results?I'm also interested in Geekbench 4 and 5 results if you don't mind running something like YABS quickly.
https://github.com/masonr/yet-another-bench-script
Basicaly just run
curl -sL yabs.sh | bash -s -- -9
Also which one is that, by 6 cores I assume VPS M?
Yeah, something like that seems reasonable enough.
But lets say if you did those tests over a long period of time and didn't get throttled at all, I'm a bit concerned about Contabo allowing, well, almost-but-not-really abuse-ish.
Then again, I've heard a lot of people just leave their VMs idling because of the high setup fees so they want to make their VPS "worth it", so the overall usage on the nodes might not be as high as I expect it to be.
Don't have it anymore
I have been using Contabo VPS S for over a year, here are my few cents:
It is ok as long as you don't expect too much. Just don't get very excited about those CPU/RAM/SSD you read on their pricing page - they are only partly "real", the other part is marketing.
CPU steal is around 20-25%, more or less all the time (with short peaks up to 30-40%). So, if you expect to have the whole 4 cores even for a short time - forget about it, you will get at most 3: https://ibb.co/Wpkckwy
Unfortunatelly, if you try to use the remaining cores with any significant load for any significant time - they will not allow that. My application was using 2 full CPU cores (=50% CPU usage on VPS S) for around a week (+20-25% steal, the total CPU usage was around 75%), then funny things started to happen - the app process was just killed, without any log records.
First I thought that the problem was with my app, but that hasn't happened anywhere else, not on my local computer, not on other VPS, only on Contabo. So I added a simple bash script for restarting the app if it stops (just a few lines of code, a simple while loop). The script itself doesn't consume CPU, it just starts an app and waits. A day or so after that the app process stopped again, the script restarted it, everything was fine. Then the app AND the script were killed! Again, without any logs or any signs of error.
My app is not memory intensive at all, it wasn't OOM kill (+there wasn't anything about OOM kills in system logs). My opinion - Contabo has some kind of backdoor in their OS images and they use it if they really have to. It's not related to virtualization, they must have direct access to your guest OS to do that. That is also the reason there is no simple way of using your own OS images ("contact our support, they will be happy to do that for additional price" - yeah, right). So much for security
/dev/sda:
Timing buffered disk reads: 1122 MB in 3.01 seconds = 373.15 MB/sec
fio Disk Speed Tests (Mixed R/W 50/50):
What happens if you write the ISO into hard drive and boot from there?
https://yoursunny.com/t/2021/SolusIO-IPv6-ISO/
The installer gets killed?
I think that 99% of all Contabo users will not bother to do that, they will just select an available OS image from the list.
Now contabo can directly choose to use your image for 0.99 euro a month, so this shouldn't be an issue anyway.
I don't care about steal which IMO is but a number that some have discovered (and misunderstood) to be the "spilling the beans" number.
Unlike most here I have had quite a few conversations with them and I've never found Contabo to knowingly tell me BS or lies. In fact, they really went out of their way to allow me to see the bloody facts. To name two particularly striking examples:
So, how to resolve the apparent contradiction? How to explain that Contabo does NOT brutally oversell - while at the same time customers report again and again flaky performance, throttling, overselling, etc?
IMO the response is simple albeit not flattering to Contabo: I'm quite confident, and have in fact multiple times observed indications to that effect, that it's a software problem. IMO they simply do not have the means in place to reasonably tightly control how, and how hard, users use or abuse their VPS. Adding to that is the fact that they run a huge operation that is, unlike a provider with say 20 or 50 nodes Contabo has no chance at all to really control all relevant factors and parameters wrt all the VMs.
Besides, and to be fair: what shall it be then, what do you want? If a (huge) provider reigns in and throttles users going beyond what is their fair share some of you make lots of noise, oh how brutally the provider throttles and limits you!!! And if he doesn't then some of you will make a lot of noise, oh how careles the provider is and how he brutally oversells!!!
Do I think that Contabo is the best provider of all? Nope, certainly not. Contabo is a huge provider selling AMD based VMs, both VPS and VDS, for a very attractive price. That's it, period. You get tons of processing power, lots of memory, relatively large NVMes (formertimes SSDs) and a copious amounts of traffic for a relatively modest price. Damn fair enough.
You want close to zero steal? Fine, look somewhere else or shell out the $$ for a VDS. You want speed demon NVMes that never go below 90%? Fine, look somewhere else.
But kindly already stop the BS campaigns against Contabo! I've rarely seen a provider playing transparent and with open cards like Contabo did. If you want, just substract 1/3 of their spec numbers and be done (and still have a not bad at all VPS).
(@AleksaDjordjic you finally triggered this post but I do not specifically address you!)
I think they're gonna be expensive if you subtract 1/3 of the "advertised specs" compared to other mainstream providers.
There are a lot of options from other providers and you are not forced to use them at all.
After reading a bunch of BS above, I can only link to this thread which I wrote up a while earlier:
https://lowendtalk.com/discussion/173716/jsg-the-server-review-king-can-you-trust-him-is-yabs-misleading-you#latest
Well, it's not that easy to get a 2.7 AMD vCores VPS (1/3rd reduction applied) with 8 GB memory, 50 GB quite fast NVMe - or even 100 GB for +€1 -, and 32 TB of traffic with decent (but not great) connectivity for €5/mo (+VAT, if applicable) ...
Again, IMO those are not "high-end" VPS (for those see -> e.g. NexusBytes / @seriesn) but they are very decent choices for many, many customers, Most customers simply do not need max. performance VM; most people just look for "decent performance (+ buzzwords like AMD and NVMe) in a no worries sized resources ("damn big enough") and a good deal" - and that's what they get.
Also keep in mind that 97+% of people don't compare against LET promos but against other big name providers and 100k+ VMs and customers seem to confirm that take.
But again my point wasn't "Contabo are the best!", nope, it was "cut the hype and bashing" plus "Contabo played hardcore open and transparent"; most providers wouldn't go so far.
They keep sending me one month free promotion, but I am not getting it.
It is not worth it, IMO. There are much better vendors.
I understand that for some people, it may be “deal of the century”, so more power to them.
"decent deal" != “deal of the century”
“decent deal” for person A could be “disastrous deal” for person B
Was this an NVMe VPS or a regular SSD VPS? I had an NVMe one in Seattle for a few months and always had less than 2% steal. I was running Sentry on it which did quite a bit of disk I/O (receiving performance data and error reports from some of my apps, and periodically optimizing/vacuuming an ~80GB Postgres database)
Not the best provider, but it definitely worked well enough! Only got rid of it because I got a $199/year dedi during Black Friday
https://yoursunny.com/t/2021/SolusIO-IPv6-ISO/
I once tried something not so different (same idea, different method). I had trouble getting it to boot after installing the OS. Also for Windows, I had huge issues with drivers that it won't even run properly. (might be me being a noob, I'm sure some people succeeded)
Got 2 VPSes in Germany. One of them have been active for over a year, the other one is new. The SSD one, the older, seem to be around 3-10% steal and the NVME one less a few %. With the NVME going up to 30% when under heavy load while SSD more or less stays the same.
But for that price I dont find it to bad to be honest, but thats my 2 cent.
The only problem I have had is at the 19th when the SSD server started to go bananas, sadly I didnt check the steal etc but opened a ticket. 4 days later I was moved to another system that have been ok soo far.
But I also noticed that I was moved from EPYC 7282 to E5-2630 v4 though I havent had any issues with such a move yet.
Im very interested in other options, in case any of the servers take the turn for the worst again.
Price below 8-10Euro
im not so sure moving from time to time is the best arrangement for production.