Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Why is there not a high load limit for more cores?
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.

Why is there not a high load limit for more cores?

Hi,

Most of the LET providers I have used or looked into do not allow a higher load limit on plans that have a higher number of cores. Regardless if one is one a 4 core plan or on a 2 core plan, often both plans seem to have the same restriction on server load.

This is confusing because, a server load of 1.0 with one core, is somewhat equivalent to a server load of 2.0 with two cores. I know this is not completely true because there are resources like IO that affects the load regardless of cpu cores. Nonetheless, one would think there would be a bit higher load limits on a with more cores because the user is renting more of the host machine. I am not suggesting a 4 core VPS should be allowed 4x the amount of load, but perhaps a bit more than the VPS plan with 1 core.

What I am missing? Or is the above too much to ask for a LEB?

Comments

  • Simple fix. Buy a dedicated server and use as much resources as you like .

    Thanked by 2doughmanes Rolter
  • sinsin Member
    edited February 2016

    @AdamM said:
    Hi,

    Most of the LET providers I have used or looked into do not allow a higher load limit on plans that have a higher number of cores. Regardless if one is one a 4 core plan or on a 2 core plan, often both plans seem to have the same restriction on server load.

    This is confusing because, a server load of 1.0 with one core, is somewhat equivalent to a server load of 2.0 with two cores. I know this is not completely true because there are resources like IO that affects the load regardless of cpu cores. Nonetheless, one would think there would be a bit higher load limits on a with more cores because the user is renting more of the host machine. I am not suggesting a 4 core VPS should be allowed 4x the amount of load, but perhaps a bit more than the VPS plan with 1 core.

    What I am missing? Or is the above too much to ask for a LEB?

    4 cores are going to usually run something at a lower load then a plan with 1 core. If I'm running something on a Ramnode KVM with 1 core and load is 0.20 and upgraded to a Ramnode KVM with 4 cores then that load is going to be lower then 0.20 with 1 core.

    Of course like you said io plays a role and then there's cpu speed, any different settings a provider has set, whether or not your app takes advantage of more cores, etc...

    At least that's the way I look at it

  • AdamM said: Or is the above too much to ask for a LEB?

    The $20-40/mo sweet spot on dedicated servers are just for you. When RTO servers started coming out and now pretty much $20-40/mo can be found at a lot of places, you'll have way more enjoyment and less stress having 100% resources on a dedicated server.

    Toss Proxmox on such a server with a /29 IP block and you've got so much more flexibility.

  • sin said: 4 cores are going to usually run something at a lower load then a plan with 1 core. If I'm running something on a Ramnode KVM with 1 core and load is 0.20 and upgraded to a Ramnode KVM with 4 cores then that load is going to be lower then 0.20 with 1 core.

    No. If you have a plan with access to 4 cores, then, the load you can have is 4.0 . If you have a plan with one load, then, the load you can have without struggling the cpu is 1.0 .
    But, you cannot push load to the limits for long periods. In fact, some providers won't agree even reach the maximum load of the cpu.
    In most cases, vps providers will allow you to have a high load (reaching the limit of your allocated cores) for short periods of time (e.g. 1-2 minutes) and have a load of 30-40% of your allocated cpu power for a little longer (e.g. 5-10 min). But they expect you not to exceed most of your server's uptime a load of 0.10 - 0.20.
    Of course, it depends on the cpu, the load of the node itself, the provider etc.
    Having access to more cores, will usually give you the opportunity to run cpu intensive tasks and complete them without crashing your vps or the allocated cpu. What can you use? It depends on what your provider means in his "fair share" policy. Something you have to ask before you buy a vps, if you think that your project maybe will need more power.

    Thanked by 1sin
  • jvnadr said: No. If you have a plan with access to 4 cores, then, the load you can have is 4.0 . If you have a plan with one load, then, the load you can have without struggling the cpu is 1.0 .

    I know that but what I was saying is with more cores then the cpu usage is going to be spread out across those cores (if whatever is using the cores can take advantage of it).

  • sb56637sb56637 Member
    edited February 2016

    jvnadr said: No. If you have a plan with access to 4 cores, then, the load you can have is 4.0 .

    But do most VPS providers actually allow you to run for extended periods of time with a load of 4.0, given this situation of 4 available cores? I was under the impression they usually don't, which I believe is the point of the OP's question.

  • @sin said:
    with 4 cores then that load is going to be lower then 0.20 with 1 core.

    Nope. Load is independent on cores.

    Load is the number of jobs in the running queue, so if you have 4 cores then you can keep up to 4 jobs in the running queue to saturate your CPU. Of course on a single core, only 1 job is enough to saturate it. For a load of 0.5, it's 1 job stays in the running queue for half of the time, or 2 for quarter, or 5 for one-tenth, etc.

    Thanked by 3sin perennate deadbeef
  • @sb56637 said:
    But do most VPS providers actually allow you to run for extended periods of time with a load of 4.0, given this situation of 4 available cores?

    No, most don't. Some do (Prometeus, for example) but they're definitely in the minority.

  • Load is one of those things that you can't really plan too well for, you could have one 8x core server with 200 containers where everyone is running stuff like web servers with an overall load of 2.00, and you could have another server with 100 containers and 40.00 load because a few people decided to run a Minecraft server with some hell hole java plugins, or even worse those firefox auto browsing bots...

    Yey for live migration!

  • @sin running something on a Ramnode KVM with 1 core and load is 0.20 and upgraded to a Ramnode KVM with 4 cores then that load is going to be lower then 0.20 with 1 core.

    I think the above is generally correct.

    @msg7086 Nope. Load is independent on cores.

    Yes, load is independent to the cores. However, since the load is the average of jobs being processed plus the jobs in the queue to be processed, a larger number of cores should have a smaller load (as jobs move quicker from the queue to the processor) - that is if my math is correct. But, of course there many factors to contribute to the load.

    Thanked by 1sin
  • jarjar Patron Provider, Top Host, Veteran

    The real truth is: almost no provider cares how much CPU you use until enough people collectively use an amount that impacts neighbor performance.

    A good provider would then say "Alright you've been using 100% of your CPU allocation for the last 4 months, we need to talk."

  • High load really sucks life out of webservers. I've got a personal kidechire running squid3,openvpn,nginx php5-fpm and mariadb, and now also runs a pastebin and librenms.

    Librenms snmpbulkwalk causes insane load on the kidechire and you can tell on your pastebin when it's polling your other servers.

    jarland said: The real truth is: almost no provider cares how much CPU you use until enough people collectively use an amount that impacts neighbor performance.

  • @jarland

    That makes sense.

  • raindog308raindog308 Administrator, Veteran

    doughmanes said: The $20-40/mo sweet spot on dedicated servers are just for you.

    The problem is the I/O generally sucks because you're talking about either single SATA or software RAID SATA at best. Lots of VPSes that will outperform a lowenddedi. However, if the OP needs to burn cores 24x7 and doesn't care about I/O then yeah, a dedi is the way to go.

    sb56637 said: But do most VPS providers actually allow you to run for extended periods of time with a load of 4.0, given this situation of 4 available cores?

    How long is extended :-)

    The problem is that eventually you reach a point where you cross into dedicated cores. If 4 vcores are assigned to you and you run a load average of 4.0 for a long time, you essentially have four dedicated cores...and that doesn't mesh with the typical LET business plan of "buy a 16-core box and put 64 VMs on it".

    While there are businesses with dedicated core offerings (AbusiveCores, WiredTree, etc.) you won't find LET hosts offering it for LET prices.

    Thanked by 1doughmanes
  • jvnadrjvnadr Member
    edited February 2016

    sb56637 said: But do most VPS providers actually allow you to run for extended periods of time with a load of 4.0, given this situation of 4 available cores?


    jvnadr said: In most cases, vps providers will allow you to have a high load (reaching the limit of your allocated cores) for short periods of time (e.g. 1-2 minutes) and have a load of 30-40% of your allocated cpu power for a little longer (e.g. 5-10 min). But they expect you not to exceed most of your server's uptime a load of 0.10 - 0.20.

  • @jarland said:

    A good provider would then say "Alright you've been using 100% of your CPU allocation for the last 4 months, we need to talk."

    This thread totally deserves a list of providers matching this policy fingerprint.

    I avoid providers with just one or two hypervisor nodes because I fear overzealous policing of load.
    Or rather those who assume customer malice, by default.

    On a related note, it really helps to stick to Pure SSD vms these days.
    Eliminating IO wait is a great way to avoid a high-load 'event' /suspension.

    Thanked by 1Junkless
  • @vimalware This thread totally deserves a list of providers matching this policy fingerprint.

    I agree! There needs to be more info about which providers have what type of policy for load. I use to be with Crissic (actually still am in a way), and he always had a very clear policy for load limits. I dislike when providers say things like: do not use high load for an "extended" period of time. How long is extended, and what is high load? And what is the result if we go over?

    My guess is that most people on LET do not want to abuse a provider. But, we want to know clear answers so we can design our systems properly.

    Thanked by 1vimalware
  • raindog308raindog308 Administrator, Veteran

    Providers never seem to have a firm policy on this stuff. I don't remember seeing ANY provider's TOS specify how much CPU you can use.

  • @raindog308 said:
    Providers never seem to have a firm policy on this stuff. I don't remember seeing ANY provider's TOS specify how much CPU you can use.

    I only recall a conversation with vpsdime support reinforcing their clearly stated load-limit +time period limits.

    It seemed like very reasonable limits at the time, although I have no memory of the numbers now.

    So, kudos to vpsdime for starters.

    Prometeus also has sane accommodation policy about load. (upto core count 's worth of load? Somebody confirm)

  • raindog308 said: Providers never seem to have a firm policy on this stuff. I don't remember seeing ANY provider's TOS specify how much CPU you can use.

    Ramnode does and Lunanode does too, there's a couple of others that I have seen but can't recall at the moment.

  • VirMachVirMach Member, Patron Provider

    CPU is probably the most "oversold" thing after maybe bandwidth. If your provider is giving you 2 cores, for example, on a 128MB RAM plan, chances are they're not running an 8 core / 512MB RAM node and splitting it across 4 people.

    They probably already factored that into the cost of extra "cores" they provide.

    So if they did 1 load per core, they might have to charge much more for a plan with extra cores, since the servers have extremely limited amounts of CPU compared to other resources.

    We charge relatively more per core on our plans, so our terms do go based on 1 load per logical core. But then that makes us look "bad" a lot of times when a customer comes on and tells us provider B is giving 4 cores on their 512MB plan and we're only giving 1. So there's different routes providers can take, and they have to have different terms and ways of measuring things based on how they decide to do it.

  • AmitzAmitz Member
    edited February 2016

    vimalware said: upto core count 's worth of load? Somebody confirm

    I have once been told by them that they are okay with a load of 4 on a 4 core VPS plan. If you use more than that for an extended period of time (they mentionned 1 hour+) and if this affects the performance of the node, then they will contact you and ask to decrease the ressource usage.

  • @VirMach We charge relatively more per core on our plans, so our terms do go based on 1 load per logical core

    Nice to know, thanks for sharing your load model.

    @VirMach But then that makes us look "bad" a lot of times when a customer comes on and tells us provider B is giving 4 cores on their 512MB plan and we're only giving 1

    That is true, hard to win here. The main thing I want to know is what a provider expects for load. No reason to complain if we know, its their house, not mine. But, would be nice if this info was readily available on provider's TOS/AUP, as opposed to the vague statements in almost all TOS/AUP.

  • I've seen guaranteed percentage of a core like 2-5%

  • @Mun said:
    Simple fix. Buy a dedicated server and use as much resources as you like .

    Until the CPU overheats and catches on fire and triggers the sprinklers where you will have successfully fried the DC entirely :3

  • @doghouch said:
    Until the CPU overheats and catches on fire and triggers the sprinklers where you will have successfully fried the DC entirely :3

    Most DCs use gas-based fire suppression before resorting to water, but I can see how funny that would be :P

  • doghouchdoghouch Member
    edited February 2016

    @0xdragon

    (also) What happens when they run out? Servers go boom-boom, am I right? :)

  • @AdamM said:
    Hi,

    Most of the LET providers I have used or looked into do not allow a higher load limit on plans that have a higher number of cores. Regardless if one is one a 4 core plan or on a 2 core plan, often both plans seem to have the same restriction on server load.

    This is confusing because, a server load of 1.0 with one core, is somewhat equivalent to a server load of 2.0 with two cores. I know this is not completely true because there are resources like IO that affects the load regardless of cpu cores. Nonetheless, one would think there would be a bit higher load limits on a with more cores because the user is renting more of the host machine. I am not suggesting a 4 core VPS should be allowed 4x the amount of load, but perhaps a bit more than the VPS plan with 1 core.

    What I am missing? Or is the above too much to ask for a LEB?

    server load of 2.0 on 2 core, is 2x more than 1.0 with one core, assuming the cores have same performance? think i'm just misinterpreting something

    but, anyway, my guess is that (unless it's very oversold) you can get much better burst performance from that 4 or 8 core VPS than from the 1 or 2 core VPS, even if the 'sustained' limit is similar

    e.g. if i'm running a bitcoind client, CPU usage won't be that high, until that new 750KB or 1MB block hits the network, then it'll require a lot of CPU usage.

    Like on my Hetzner server;

    2016-02-05 09:25:28.974879 - Connect block: 39.25ms [90.49s]

    .... on some single core VPS, this would probably be 500ms+ ... on an 8 core VPS, probably just a bit slower. The sustained usage would still be low

  • @Amitz said:
    I have once been told by them that they are okay with a load of 4 on a 4 core VPS plan.

    Yeah, that's what I've been told too and what Mao's said on LET:

    @Maounique said:

    the 100% load clause applies like this:

    1. For budget plans, you can never exceed the maximum load (as in the example above, 4 for a 4 vcpu plan). You can stay at exactly 100% but never go above, which it means you should consider spikes in your apps and use conservative settings.
    2. For regular (pro) plans, excluding budget brands, such as overzold, xenpower, you can go over 100% for limited time if needed, such as a reddit of your page, a sudden jump in visitors, compilations and streaming demand, etc. We have the extra capacity for this.
    3. For Biz plans, you can often and for a long time go over the limit, however, more than 200% is not accepted unless in short bursts, like half an hour or so, over 1k% is not allowed, period. That usually means hacking or major malfunction which could lead to data loss so we shutdown and give you a notice.
Sign In or Register to comment.