All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Reasonable CPU Use: A Look at AUPs
A couple threads here had me wondering what providers/users think is reasonable use of a VPS in terms of consumption of resources. Let's assume that using all of one's bandwidth and disk space is a given, and I think it'd be hard for a provider to complain if you use all the non-burst RAM you're assigned. So we're really just talking about CPU and I/O load.
For the cloud big boys (Amazon, etc.) I can I expect I can launch a recursive computation of pi to the final digit while converting my tentacle hentai collection from 5K to 320x240 because they throttle, etc.
I think the next tier ( Linode, DO, Vultr ) doesn't care either...at least, I've run some heavy compute stuff on DO for example without complaint.
But what about your typical LEB host with their Solus-admin'd OpenVZ? What's reasonable for a typical provider?
I was curious so I did a little sampling. Roughly from "most specific" to "most vague"
Virmach
High Usage Policy.Any usage disrupting the overall performance of our server(s) is not permitted. However, at our discretion, we will generally follow the guideline below: High CPU: Customer’s Service (1) cannot burst to 95-100% usage for more than 5 minutes, (2) cannot average higher than 50% usage within a 2 hour period. Gaming plans, services with the high CPU option, and any custom high CPU usage services previously discussed with VirMach may burst to 100% at all times. High Load: Customer’s Service (1) cannot have a 15-minute load average higher than the number of full logical cores assigned, (2) cannot have a 1-day load average higher than 70% of the number of full logical cores assigned. High Mail Volume: Customer’s Service (1) will have port 25 blocked by default, (2) may send 100 maximum e-mails per hour, (3) must maintain a similar average volume of mail on a week-to-week basis – no bursting. A different arrangement (including some services) may be explicitly discussed with VirMach to override this policy. High I/O: Customer’s Service (1) cannot have an average IOPS of more than 80 within a 2 hour period, (2) cannot burst to 300MB/s or more disk write average for more than 10 minutes, (3) cannot have more than 300 write operations per second average for more than 1 hour, (4) cannot be above 20% average utilization within a period of 6 hours. High Network Usage: Customer’s Service (1) cannot have more than 50,000 conntrack session at any given time, (2) cannot use more than the allocated bandwidth. Customer understands that the network is shared and utilizing maximum network speed may not always be possible.
In summary, you may buy the VPS but usage is frowned upon.
SecureDragon
"VPS CPU AND LOAD POLICY • VPSs with a load above 8.00 will be automatically restarted by our automated system. • VPSs that maintain a load average above 0.90 for more than 4 hours will be automatically throttled to 50% of their CPU core until a ticket is opened by the client for review."
Clear rules, easy to understand.
Ramnode
If your VPS is consistently maxing out one full core or more and impacting the performance of other client servers sharing the same host node, we may ask you to reduce your usage."
We're getting a little vague now, as "consistently maxing out" is not as precise as SD's load average number.
The following are all too vague to really understand if your usage could might the abuse criteria:
BuyVM (presumably not their dedicated slices)
"Resource Abuse consists of any activity, intentional or otherwise, that consumes sufficient system resources to negatively affect other clients or equipment."
Hostigation
I didn't find anything.
CatalystHost
Server Abuse Any attempt to undermine or cause harm to a server or customer of Catalyst Host is strictly prohibited. As our customer you are responsible for all your accounts. This includes abuse of the server’s CPU and RAM resources to a degree that effects the performance of other customers on your node.
Prometeus
Prometeus in its sole discretion may discontinue service without any notice to any Hostiing, VPS or iwStack customer that uses a high amount of server resources (such as, but not limited to, CPU Time, Memory Usage, and Network Resources)
I think these provider's actual acceptable use is "you can use as much cpu as you want as long as no one complains" but it's hard to know as a user what that allowed usage is.
Comments
Rule of thumb is if they don't explicitly advertise unlimited or dedicated CPU on sale/pricing page, find another provider (I don't bother navigate further to AUP page).
IMO the acceptable load percentage mentioned in AUP is just approximate number. Even if you don't reach the criteria but they believe you're using too much resources, they will likely warn or suspend you.
Interesting.
I think the maximum acceptable load should be around the number of cores available. But still, that may be too high...
Not enough for me. I mean, that load, on the 1min, 5min or 15 min? For how much time > 8.0? At the first indicator it will be restarted inmediately? Because maybe a little spike at the 1min cpu load mark should be ok.
That's interesting.
No Dewlance review???
It's called AutoBoot, not AutoCPU.
Francisco
VirMach's AUP doesn't seem too bad for personal use. I understand why one might not want it for business use, though.
>
Our system checks the 15 minute interval, if it's at 8.0 or higher it gets automatically rebooted. We implemented this when we were getting hammered with SSH attacks that were overloading our nodes. I just checked and the last time a user's VPS was restarted for hitting this limit was a few months ago when a client's VPS started spawning Apache workers like crazy.
Our AUP specifies load because in Linux load can be caused from both CPU and disk IO, neither of which are dedicated resources so this is a fair way to monitor both resources.
I think you mean 'hot'.
Good idea.
No, you can't max out anything with ramnode. Receive one ddos on their openvz vps that causes load and that their ddos filtering doesn't detect and you're most likely terminated or your vps gets rebooted all the time.
On the kvm plans however you can abuse CPU as you like.
Speaking of digitalocean: You can't max out the allocated bandwidth, you will get suspended.
Linode is probably the exception with me using 3 of their vps to download at 1.2gbps for a month.
This is really something I am think recently, as Sentris thing happened to me.
I made this comment few hours ago.
I understand why providers have to be careful and serious about specific user's high cpu usage. I just don't agree with those who terminate service because of this. That's really make me feel the word, bait...
As your examples above, SecureDragon and Ramnode are good examples. They reboot boxes, set up limits and communicate with people, that's how decent businessman do.
Virmach is, from my point view, at this aspect, a bad example. I remember they have the term that terminate any abuser without notice (let know if I am wrong). With that kind of limits, first question is: even that is cheap so much, but what it can be used other than idle? second is: why just terminate them but not reboot them or shut down them? if the purpose is to avoid cpu abuse but bait people. third:a conspiracy thought, how do I even know if I did use that much cpu? Maybe a provider just set a very low and unfair limit to find excuses. Or is this how some company can consistently provide cheap boxes regularly? For example Sentris and ...
Of course, those well-known and reliable providers did give out some sort of similar boxes once, for example, ramnode, hostus @AlexanderM, tragic server @tragic. I'd prefer to believe that is kind of business strategy. Because these guys raise price after a period to a sustainable level, although they give out some cool offers, but may not be able to compare with themselves before. The most important, they keep providing excellent service, no matter when you became their customer or how much did you pay!
This is one of those things where I've always preferred an unstated policy. I know it's weird, but it's where I'm at. The reason is simple. If you oversell, and I want my provider to oversell (because I like good prices), calculating the CPU down to the amount you would have to assume dedicated (meaning the max one server could cap if all servers capped equally at the same time without causing issues), you come out with a number that I am never positive that I can stay under.
This has me uncomfortable and feeling uncertain, like things outside of my control could add a 0.01 load value for 3 seconds too long and I could be suspended. How the hell am I supposed to plan ahead for server load values and times to those specific values?
I know that a number always exists. There's always a number of "X containers can sustain Y load simultaneously before a problem occurs." The thing is, that situation is just so far fetched and theoretical that I'd rather my overselling provider take on the risk and responsibility than to pass it on to me. If I'm being a complete dick with my server load then that's another story, hold me accountable.
I know bandwidth can be said the same way, in that DDOS is not usually your fault and you can't always plan ahead for it, but I'd rather risk a DDOS at an unprotected provider than be held accountable for a 0.05 load increase for 6 minutes. Probably because a DDOS is more problematic than a VPS maintaining a load of 3, and I know that.
My feeling is also probably similar to that of someone choosing unlimited shared hosting. We all know limits exist, we just don't want to think about it. Choosing a provider that implies that they'll take on the risk of the possibilities that are within reason, instead of demanding that I constantly count grains of sand and holding me accountable if my scale is one grain of sand too heavy, makes me sleep better at night.
@jarland
cannot Agree more
Seems acceptable
So I actually looked into this because I'm interested in their services (haven't purchased one yet but the reviews seem good and the deals seem good too) and I don't mind reading AUPs or ToSes.
I don't know about your questions but they don't terminate any abuser without notice.
Anywho, here's from their AUP about abuse:
So it seems like it's not an instant termination unless you were doing something with prohibited content (and they specify what counts as prohibited content).
Hope this helped you.
This thread has certainly made me question purchasing with them in a neutral way.
yeah it helps
Thanks OP for saving me an hour at least.
I was thinking of doing a comparison for myself before making recommendations to others.
Edit: If someone wants to make a comparison of vps providers with nullroute/autonull plus the nulling policies , I would be grateful.
(Autonull is perfectly acceptable for websites since attackers usually pay per minute or hour on booter sites.)
We have pretty heavy CPU restrictions (shutdown after 3.0+ load) just due to the nature of abuse we've received in the past. I would love to remove this, trust everyone and we're all a happy family. The reality is not everyone feels this way and just doesn't care.
I'm also pretty flexible. Happy to add and have added customers on a whitelist if they are actually using a reasonable load.
It's always ok for most legitimate users if you are ready to hear their reasons
In my books if provider is willing to be flexible I will mirror the same.
60% of a core avg over a 24 hour period is my line, and that is only enforced if it is impacting others.
Disk IOPS over 300 p/ second avg over. 1 hour you get a warning, over 600 you get shut down (not suspended), you get 3 chances to sort that out then your gone.
Do such limitations scale with vps size? (ie 2 core vps, 1.2 load over 24h?)
What about IOPS and bigger vpsses?
That's another dimension I hadn't thought of...I was thinking bandwidth in terms of GB/mo but you're talking rate of consumption.
I'd assumed providers throttled network with QoS settings, so you can use 1gbps if you're the only guy talking at that moment, but usually you're sharing your percent/burst with everyone else. In other words, I can say "application, use 100gbps" but I'm only going to get the rate the provider allows, up to my bandwidth cap, and networky QoSy things keep it all in place with policy.
No?
So what if you have a 4-core VPS? You could never use all four simultaneously...? Also, is that 3.0 immediately (going from 2.99 to 3.0 results in a shutdown) or over a period of time? Just curious, as, well, I have a 4-core VPS with Tragic :-)
It's over a period of time. If you need to burst, we usually just whitelist (so just ticket in). We've found that most abusers get shutdown and just ignore their VPS after.
IOPS does not scale... because.. it just does not scale
CPU, yes, for every 2 cores you can use 60% of 1 for 24 hours avg.
This is why I prefer dedicated servers for heavier workloads - yeah, I might be able to get something a little cheaper, but it's worth paying the extra not to have to think about it.
Any proof of this?
According to this https://clientarea.ramnode.com/knowledgebase.php?action=displayarticle&id=52.
I don't think you can max out CPU on KVM.
@jarland - is this true? Are you not allowed to use your full allocation of bandwidth?
Digitalocean does not count traffic at all, so what do you expect? Can you use 1gbps on their 5$ plan all the time?
No you can't:
This happened after continuous use of 1gbps download after some hours.
You can't abuse a KVM vps as much as an OpenVZ VPS cpu-wise, you also do not cause that much troube for others. While you can cause cpu-load of 100+ on an OpenVZ VPS and this will directly translate to the host node, on a KVM VPS a load of 100 will probably not cause a load of 100 on the host node.