All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
What are your feelings on Host CPU throttling when you're using less than 25% of the core?
Hi Folks.
I've got a couple cheaper legacy OVZ which I've been careful to not intentionally abuse- network, or otherwise. I've even setup scripts to ensure that I never burst (much) over 20% load, and do my best to be a "good neighbor". That's just polite on a shared service.
Even when holding my stuff down to around .2/core (.2*NUMCORES=x, so, a 5 vCore system would be a load of 1.00- _or less_ ), I still have a couple OVZ providers that tune this down to LOWER THAN THIS -so my processes fail to run, or are exhausted after a few minutes. These aren't $3/yr NAT services, either.
I've read through SLAs, and it's not like I am not trying to mine buttcoins.
What are your feelings on a provider who only wants you to idle and not actually use your services?
I'm not a guy who generally asks for refunds, but if you don't want me to use your service- why would you rent it to me in the first place?
- Really, does it?130 votes
- Yes46.92%
- No12.31%
- $712.31%
- Debian, thx.  7.69%
- ServerHand15.38%
- VolumeDrive  1.54%
- Node.js  3.85%
Comments
Bursting all cores for a short period of time should be possible.
If not what's the point? The provider may as well clocked the virtual CPU.
By your own words, DICKS?
It's not even bursting- it's a semi-sustained under .2 load. If it goes over .2, I stop it, and continue the process when the loadavg normalizes- which is actually better than cpulimit offers, as it's load based, rather than CPU only. I've never seen more than ($NUM+.05) in my monitoring ($NUM +/- .05).
Pretty close to how I feel about it, too.
I'm going to go with "it depends". We have automated systems in place but depending on the client and the usage we've been known to bypass the throttling for clients. We also make sure that our system opens a ticket to inform the client so they can open a dialogue with us. 99% of the time when a client gets the notification they open a ticket with "what is causing this load?" and we help them get their poorly optimized script or hacked VPS under control.
@KuJoe This is "no contact, just processes stopped". Obviously, if someone asked "WTF ARE YOU KEEPING A .1 LOAD ON FOR?", I'd have explained that it is just a stupid LET thing.
Plus, you haven't bitched at me for my existing load.
What if I told you that one is a well respected member of LET, and the other recently acquired one of the same?
Did you contact with him? If yes, what was his response? (genuine interest)
Not yet. I wanted to make sure I didn't have some crazy perspective because I am paying peanuts- but not really using the services I paid for. (Cents a day, which is common for LET, specials and not.)
Well, a load of 0.2 is not much and this should be also stated to AUP/TOS. But, for a service of , let's say, 5$ per year, a provider will stuff the server with 100's of vps and he is hoping that almost all of them will be idle most of the time...
Anyway, cents a day is definitely more than $3 per month, so, it's not too cheap not to allow a 0.2 load. On the other hand, if they give you 5 cores and you have 0.2x5=1.0 load 24/7, then, this could be taken as an abuse of the service. But IMO, a provider should write in his AUP/TOS what is the maximum load/time allowance in his servers.
So far, only one host (a new one to me) has said "Dude, you were blowing over .75, and you can't have that"- and in my defense, it was a runaway process. I fixed, they closed ticket, life goes on.
This is actually a 6 core CPU and it isn't a guaranteed hard 1.0 load - I just force it to stop if it hits that. Most of the time it's less. I agree about the AUP/TOS - but to not even contact me before locking down the use is pretty shitty, IMO.
In this case, it's not a good policy from the provider...
I would suggest to contact the host. Only him can give a real and exact answer
Maybe that will be our next service, cheap VPSs with a max load of 0.1 and 10 IOPS. I see idleservers.com and idlevps.com are available.
Looking at the loads on our VPS nodes now, I can see some good candidates for such a service.
Cheers.
Dave? English fail? Are you OK?!
That is less acceptable than bursting, imho. If you're on a basic 128MB openvz running on an 4-core 32gb e3 box, you have 1/256th of the memory before any oversell so you shouldn't expect much more than 1/256th of the cpu, i.e. 1/64th of a core or 1/32th of a hardware thread, sustained. That said, bursting should be allowed.
Tiredddddddd, neeed to find coffeeeeee...
As I said, I'm not really doing anything magical here. In this instance, I'm sitting on a 6GB 6 vCORE system with a load under 1, and that's the highwater mark. Like I also said, I'm not running buttcoins or anything silly like that. Just, if I need to keep my own personal load below 0.09, why do I have the service?
On the other hand, I've been beating my NetCup threads with no remorse for LET's RC5 project, because they're doing little else. No shits are given, since my load is under 1.
Add some Baileys. Much better that way.
NO! NO! NO! AND NO!
We allow on ALL services, including OVerZold, 100% usage, i.e. load of 1 for one core, 2 for 2, etc. On Biz/Pro plans, you can go way more than that, as long as it is not permanent or not 10x, i.e. load of 40 for 4 cores, abuse means you got over 100% on budget services (some leeway is acceptable, like 110% for half an hour or 200% for a few minutes) or you have sustained loads of 200% or more on premium services.
Even more, we do not check, even, if the load on the node is acceptable, only if the dashboard says load is approaching 80-90% which shouldnt happen, we stay below 50% normally, we go in and check wtf is going on and that is usually some hack or mistake, like a cron running amok or something.
On the other hand, yeah, we dont have any product at 7 USD a year for 4 GB ram, that is monthly price here, so, maybe, you are right and people had too big an expectation.
IMO, instead of losing time policing users, tickets, marketing to replace the ones lost, fraud because price is too low and people treat the VM as a throw-away, we can save that time and offer a product with some serious margins before saturation occurs.
This way, people know it is hard to get in and try to follow the rules, they respect us, we respect them.
A 6 vcore should have full use of load 1.2+. Otherwise what the hell is a 6 core. lol.
OpenVz & KVM should play by the same rules. They both can be oversold just a matter of ethics and marketing matching some semblance of fair-usage. Bursting at full core should be realistic.
I guess it comes down to what you pay as well. Caveat Emptor.
Provision 100 OpenVZ containers on 4 cores, allocate 2 cores each. Refund top CPU usage customer and send them on their way. Enjoy idle server.
^ Playbook for guaranteed success that doesn't involve throttling burst users. Just kindly part ways with the one customer out of 100 that actually uses any consistent CPU at all and win.
It all depends on the terms provider have for his services. A $5/y 512 OpenVZ vps with 5 vcores could allow spikes of, let's say, an hour of 100% usage, but no more than 10% constant usage of the allocated cpu power. IMO, it should be cleared on the AUP/TOS, though.
And, c'mone with your overzold service. The name is a scam! It's fraud against all respected LET providers that are actually do oversell their crappy vps services!
I do not have active overzold service, but for the period I had one, the performance was superb, far away from a really oversold server.
You are disgusting, you create a bad name for the term "oversold"! You claim it is oversold but compared to real LET providers, it's not even a 10% oversold! Shame on you!
For LET VPS, I would say 25% on average, and 50% burst max 2 hours should be allowed, Although I had seen some higher limit in some TOS but I don't think the host could afford that.
One question, how does cloud system like AWS, GCE, Azure or Aliyun handle CPU stealing so well without crashing the processes?
Please cease using this as a reference point, as I've already stated that this is not the case.
100% of @jvnadr shitposts vs 150% of WSS shitposts? Same difference and similar timespan, but the timeframe metrics? Hardly the same.
It's not about your example, it's an answer to Maounique about the perspective some providers they may have. What I really wanted to say, even if I wasn't so clear, is that if they do have such a policy for such a service (as an example), they should state that clearly in their TOS/AUP.
Don't be so WSS egocentric!
DIcks, floppy dicks.
But, but, but... MY THREAD!
Imma gonna go to my safe space and play in a coloring book until my personal anguish drops below 1.0!
Apologizing AND using one of your favorite words! Nice!
Actually, it is oversold, some 3-4 x depending on node, BUT, people like their service as it is usable, good performance and not too expensive, some 5 GBP per xe.com now:
5 GBP =6.70284USD so it would fit LET prices for 4 GB ram, 4 cores and 50 GB disk.
As they dont want to lose it, dont abuse it, dont pay with stolen money and dont chargeback either, so we can afford a good server and not charging an arm and a leg either.
ATM we have 4 servers with OVerZold, it is a current (not legacy) product and is available for everyone, unlike our 12 or 6 Eur a year VMs which are only available for existing customers as we do not wish to risk abuse for pennies on our Business network.
fuck the police
max all cores 24/7 yo