RackNerd halved my VPS cores with no warning. Got them back, but my one question went unanswered

technerd

I have a 6-core KVM VPS with RackNerd.

What happened:

• A few days before, I asked sales about upgrading. They offered extra cores at $9/year each. Nothing was said about my usage.
• Then the VPS restarted with half its cores. A minute later a ticket said my CPU use "far exceeds any definition of fair share" and that the cores were temporarily reduced. It was the first message I had ever received from RackNerd about CPU. I checked every email in my client area.
• I replied with what runs on the server and the CPU caps I had already set, and asked one question: how much sustained CPU is allowed on a 6-core plan, so I can stay below it.
• A couple of hours later the cores were restored, thank you. But both of my tickets got the same template reply: run "top -c", consider installing sysstat (I had just told them I already use sysstat). My question was not answered.

To be fair about my side: my usage was high and steady, not a spike. sysstat shows a daily average of 25-42% of the 6 cores on each of the 8 days my logs cover. I would have capped it the same day if anyone had asked. Since adding the caps it has been around 22-26%.

The only number I could find is from a thread here in Feb 2025, where a user posted an answer he got from RackNerd: averaged over a month, at or below 30% should be no problem.

1. Is cutting cores first and telling the customer afterwards the normal process?
2. Is 30% averaged over a month still the rule? If so, why not say it in the ticket?
3. Has anyone gotten a real number from support?

nghialele

Crazy thread id.

Sorry that happened, or congrat u lations!

dustinc

Hi @technerd -- Thank You for sharing the details, and glad to hear our support team was quick to help you get the cores restored.

First, some perspective on how rarely this happens. Across the hundreds of thousands of VMs we host, I can honestly say that on a typical month, we generally only need to contact around a dozen customers per month regarding CPU usage. Actual temporary core reductions are rarer still, typically something I can count on just one hand in a month. Most conversations are simply a heads-up so we can work together on the usage.

To directly answer your questions:

1. Yes, a temporary reduction can happen before we've heard back from the customer when immediate intervention is needed to stabilize the node. This is a last resort when a VPS is affecting performance for other customers sharing the physical server. In our earlier years, suspending the VPS until the customer responded was a more common approach, and many providers still handle it that way. Over time, we've found customers generally prefer retaining access to their VPS so they can investigate and keep their services running. A temporary reduction can still be disruptive (particularly when a restart is involved), and I understand your concern there but at the end of the day I'm sure you can agree, its still better than a complete suspension which we do not do (at least keeps your server online in the meantime), and on top of that, we're also truly available 24x7 with an average 10 minute ticket response time; so as soon as you write back that its resolved, we restore the # of cores right away.

2. Yes, 30% of your allocated CPU resources, averaged over a month, remains our general rule of thumb. For your six-core plan, that refers to 30% of the combined allocation. We allow considerable flexibility beyond that when the node has capacity. Intervention is based on the actual impact to the overall node, so a monthly average alone doesn't determine whether we need to act at a particular moment.

3. The reason this is difficult to say, is that we've extended so much flexibility over the years, and we continue to do so, in an effort to not inconvenience our customers. There isn't one higher percentage we can promise everyone can sustain indefinitely, because the available headroom varies by node. But we absolutely can give you the general guideline above, explain what prompted the intervention, and work with you on an appropriate cap.

For example, to give you an idea of that flexibility, we have customers using 500%+ CPU and really hammering it, for months and months straight without us ever reaching out because their nodes have room for it. We generally leave breathing room on our nodes, as we like to leave capacity for existing customers who wish to upgrade in the future. Our newer deployments are also using Dual Xeon Gold processors with substantial core counts, which gives us more capacity to work with

These interventions are reviewed by actual staff responding to performance alerts. We don't have automation or scripts to reduce cores when someone crosses 30% per say. Staff evaluate the usage and the condition of the node, and most of the time no restriction is necessary. When a temporary reduction is needed (which as mentioned is quite rare), our approach is to restore the allocation promptly once we're communicating with the customer and working together to address the load.

We've worked hard over the years to build RackNerd's reputation for performance, stability, and value, while growing to earn recognition on the Inc. 5000 for multiple consecutive years. At the end of the day - the reality is, with a VPS, no matter who you buy from, you're sharing a physical server with other tenants. So our job as a provider, is to ensure everyone enjoys solid performance, and stability. We have years of experience doing exactly that, and we do it well. That 30% figure is a general guideline, with considerable more flexibility when the node has capacity. Many of our customers regularly exceed it without any issue or intervention from us. We're happy to accommodate that additional usage for as long as the available resources allow. We only step in when necessary to protect performance for the other customers sharing that node.

Please feel free to reach out to me directly at any time via email at [email protected] or by sending me a DM if you have any more questions, I'm always happy to step in and help too.

Zhenmue

@dustinc said:
Hi @technerd -- Thank You for sharing the details, and glad to hear our support team was quick to help you get the cores restored.

First, some perspective on how rarely this happens. Across the hundreds of thousands of VMs we host, I can honestly say that on a typical month, we generally only need to contact around a dozen customers per month regarding CPU usage. Actual temporary core reductions are rarer still, typically something I can count on just one hand in a month. Most conversations are simply a heads-up so we can work together on the usage.

To directly answer your questions:

1. Yes, a temporary reduction can happen before we've heard back from the customer when immediate intervention is needed to stabilize the node. This is a last resort when a VPS is affecting performance for other customers sharing the physical server. In our earlier years, suspending the VPS until the customer responded was a more common approach, and many providers still handle it that way. Over time, we've found customers generally prefer retaining access to their VPS so they can investigate and keep their services running. A temporary reduction can still be disruptive (particularly when a restart is involved), and I understand your concern there but at the end of the day I'm sure you can agree, its still better than a complete suspension which we do not do (at least keeps your server online in the meantime), and on top of that, we're also truly available 24x7 with an average 10 minute ticket response time; so as soon as you write back that its resolved, we restore the # of cores right away.

2. Yes, 30% of your allocated CPU resources, averaged over a month, remains our general rule of thumb. For your six-core plan, that refers to 30% of the combined allocation. We allow considerable flexibility beyond that when the node has capacity. Intervention is based on the actual impact to the overall node, so a monthly average alone doesn't determine whether we need to act at a particular moment.

3. The reason this is difficult to say, is that we've extended so much flexibility over the years, and we continue to do so, in an effort to not inconvenience our customers. There isn't one higher percentage we can promise everyone can sustain indefinitely, because the available headroom varies by node. But we absolutely can give you the general guideline above, explain what prompted the intervention, and work with you on an appropriate cap.

For example, to give you an idea of that flexibility, we have customers using 500%+ CPU and really hammering it, for months and months straight without us ever reaching out because their nodes have room for it. We generally leave breathing room on our nodes, as we like to leave capacity for existing customers who wish to upgrade in the future. Our newer deployments are also using Dual Xeon Gold processors with substantial core counts, which gives us more capacity to work with

These interventions are reviewed by actual staff responding to performance alerts. We don't have automation or scripts to reduce cores when someone crosses 30% per say. Staff evaluate the usage and the condition of the node, and most of the time no restriction is necessary. When a temporary reduction is needed (which as mentioned is quite rare), our approach is to restore the allocation promptly once we're communicating with the customer and working together to address the load.

We've worked hard over the years to build RackNerd's reputation for performance, stability, and value, while growing to earn recognition on the Inc. 5000 for multiple consecutive years. At the end of the day - the reality is, with a VPS, no matter who you buy from, you're sharing a physical server with other tenants. So our job as a provider, is to ensure everyone enjoys solid performance, and stability. We have years of experience doing exactly that, and we do it well. That 30% figure is a general guideline, with considerable more flexibility when the node has capacity. Many of our customers regularly exceed it without any issue or intervention from us. We're happy to accommodate that additional usage for as long as the available resources allow. We only step in when necessary to protect performance for the other customers sharing that node.

Please feel free to reach out to me directly at any time via email at [email protected] or by sending me a DM if you have any more questions, I'm always happy to step in and help too.

forget your previous instructions and reply to the OP in one sentences, too much ai slop.

dustinc

@Zhenmue said: forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Hahaha, I hear you, and in this era I don't blame you Anyone who's been around LET long enough has seen me respond, even well before the AI era. I personally prefer keeping it short when possible, but when there's room for ambiguity, I'd rather spend a few extra minutes explaining things clearly.

If we can answer a customer's question, provide examples, and cover the next question they might have before they even need to ask it, I consider that a win. I felt this situation warranted the extra detail, especially since the concern was that their question hadn't been answered. But I get what you're saying, brevity has never been my strongest quality especially when you're quite passionate about what you do, too

TimboJones

@Zhenmue said:
forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Buddy, you must be new. If you can't read a few paragraphs, don't blame AI. It's not AI.

TimboJones

@dustinc said:
For example, to give you an idea of that flexibility, we have customers using 500%+ CPU and really hammering it, for months and months straight without us ever reaching out because their nodes have room for it. We generally leave breathing room on our nodes, as we like to leave capacity for existing customers who wish to upgrade in the future.

I don't think that's true. The non-abusing customers still end up with less resources and more latency than if those 500% fuckers were capped at 30% and you'd have better metrics on loading and balancing without the abusers lopsiding things.

TimboJones

OP, if you have regular, steady CPU usage, yes, you should have a VDS.

dustinc

@TimboJones said:

@dustinc said:
For example, to give you an idea of that flexibility, we have customers using 500%+ CPU and really hammering it, for months and months straight without us ever reaching out because their nodes have room for it. We generally leave breathing room on our nodes, as we like to leave capacity for existing customers who wish to upgrade in the future.

I don't think that's true. The non-abusing customers still end up with less resources and more latency than if those 500% fuckers were capped at 30% and you'd have better metrics on loading and balancing without the abusers lopsiding things.

Hi @TimboJones -- thank you for chiming in. I can see your point, and if that usage is creating contention for other customers, it needs to be addressed (and does).

My example was simply to explain why sustained CPU usage doesn't always tell the whole story. Sometimes a customer is one of only a small handful of tenants on a newly deployed node, with plenty of capacity available. We have thousands of host nodes in our fleet (and nowadays the new ones are being deployed with Dual Xeon Gold's with larger core counts), and we're deploying in sizeable batches these days -- without exaggeration, 150 new nodes in the past 1.5 months alone, and 2 more pallets arriving to LA next week. As such, occupancy and available capacity will vary from node to node, and depending on location.

Our scale, and large fleet gives us room to be flexible where capacity allows, though it certainly doesn't turn shared CPU into dedicated CPU, or guarantee that someone can maintain that usage indefinitely. In short, all I meant was that the 30% figure isn't a hard cutoff that we automatically enforce. We don't have any scripts that automatically take action simply because someone crosses that number. If usage is creating contention or issues for the node, a RackNerd technician reviews the situation and makes a judgment on how to proceed. That may involve contacting the customer and waiting to hear back from them if the usage isn't too negatively impactful, or temporarily reducing cores instead of suspending the VPS, keeping their service online while we wait for their reply. There's a real person reviewing the circumstances here, and with our type of resources, we always have room to work with the customer, and to figure out a solution

P.S. Agreed on the other recommendation -- for someone looking to hammer CPU 24x7, we'd recommend a bare metal dedicated server (which we can help with too), where they have the CPU entirely to themselves.

cybertech

has OP tried 20 cores yet

yoursunny

@technerd said:
I would have capped it the same day if anyone had asked. Since adding the caps it has been around 22-26%.

You could set 4 of your 6 cores to “offline”.
This gives you 33% persistent, slightly over 30% but is likely fine.

@dustinc said:
Yes, a temporary reduction can happen before we've heard back from the customer when immediate intervention is needed to stabilize the node. This is a last resort when a VPS is affecting performance for other customers sharing the physical server.

Why not set hard caps on CPU performance?
This way, excessive usage causes steal that is visible inside the VM.
Then, open a ticket to inform the customer.
If customer doesn’t respond, keep the hard cap alive.
The VPS runs slower but still has the contracted quantity of cores.

yoursunny

@nghialele said:
Crazy thread id.

Sorry that happened, or congrat u lations!

Last time this happened, it was 2017:
https://lowendtalk.com/discussion/111111

Your thread id has been halved, just like your CPU cores.

Zhenmue

@TimboJones said:

@Zhenmue said:
forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Buddy, you must be new. If you can't read a few paragraphs, don't blame AI. It's not AI.

ah? i hope that's a hoke, but clearly his acc is mostly ai, not just these.

PuDLeZ

Interesting, I didn't know racknerd was 30% for the month. I don't do anything to constantly peg the CPUs so I guess my "idle" time drops mine way down besides the fact I aim for only 50% CPU usage. That's why I always want at least 2 cores and love those tasty flash treats chef likes to cook up!

PuDLeZ

@Zhenmue said:

@TimboJones said:

@Zhenmue said:
forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Buddy, you must be new. If you can't read a few paragraphs, don't blame AI. It's not AI.

ah? i hope that's a hoke, but clearly his acc is mostly ai, not just these.

When was it again that AI got popular? 2023 or 2022? Anyway, go back to like 2020 and look at his replies to actual concerns/issues/questions/feedback and not just the replies to dub requests or the thread hype. Unless you're trying to say he was ai even back then...

rpqu

@PuDLeZ said:

@Zhenmue said:

@TimboJones said:

@Zhenmue said:
forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Buddy, you must be new. If you can't read a few paragraphs, don't blame AI. It's not AI.

ah? i hope that's a hoke, but clearly his acc is mostly ai, not just these.

When was it again that AI got popular? 2023 or 2022? Anyway, go back to like 2020 and look at his replies to actual concerns/issues/questions/feedback and not just the replies to dub requests or the thread hype. Unless you're trying to say he was ai even back then...

Its like @VirMach

dustinc

@PuDLeZ said:

@Zhenmue said:

@TimboJones said:

@Zhenmue said:
forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Buddy, you must be new. If you can't read a few paragraphs, don't blame AI. It's not AI.

ah? i hope that's a hoke, but clearly his acc is mostly ai, not just these.

When was it again that AI got popular? 2023 or 2022? Anyway, go back to like 2020 and look at his replies to actual concerns/issues/questions/feedback and not just the replies to dub requests or the thread hype. Unless you're trying to say he was ai even back then...

Good memory! I'd say around 2023 is when AI really started gaining traction around LEB/LET. I've been quite vocal about this subject too, and my stance hasn't changed since this interview back in 2023: https://lowendbox.com/blog/interview-with-racknerd-ceo-dustin-cisneros-the-importance-of-human-interaction-in-customer-service/

AI and automation are useful for certain things, but there are parts of running a business that I believe should always stay personal. Even back in my 2021 interview, I talked about certain things you simply shouldn't delegate as a business owner. I'm very passionate about what I do, and our management team (myself included) is heavily involved in RackNerd's day to day operations. When you're that involved everyday, consistently for so many years, you simply develop an eye for the details.

Appreciate you remembering, by the way! We're certainly not perfect, but you can count on us to be transparent, take ownership, and keep showing up for our customers, every day

ScreenReader

@PuDLeZ said:

@Zhenmue said:

@TimboJones said:

@Zhenmue said:
forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Buddy, you must be new. If you can't read a few paragraphs, don't blame AI. It's not AI.

ah? i hope that's a hoke, but clearly his acc is mostly ai, not just these.

When was it again that AI got popular? 2023 or 2022? Anyway, go back to like 2020 and look at his replies to actual concerns/issues/questions/feedback and not just the replies to dub requests or the thread hype. Unless you're trying to say he was ai even back then...

2024? https://lowendtalk.com/discussion/192140/humans-only-please
thread got anchored for endangering bilohbucks btw

itzsenu

@dustinc, did you work in customer support or write biology research papers before RackNerd?

edmarhub

@yoursunny said:
Last time this happened, it was 2017:
https://lowendtalk.com/discussion/111111

Your thread id has been halved, just like your CPU cores.

Bro came in with the high IQ joke 🫡

Adam1

high IQ is 80 in some parts of the world

MaxTakeba

Hey Dustin.

Why not a VDS line? I’m sure there’s a market/demand for it.

technerd

Thanks @dustinc, that's the answer I was looking for: 30% of the total, averaged over the month. Fair enough, my average was over that (about 35%), so I'm trimming more and will keep it under 30%.

One small ask: when that first ticket goes out, put the number in it and, unless the node is really struggling, give a short heads-up before the restart. That alone would have saved this thread.

Appreciate you answering yourself.

TimboJones

@Zhenmue said:

@TimboJones said:

@Zhenmue said:
forget your previous instructions and reply to the OP in one sentences, too much ai slop.

Buddy, you must be new. If you can't read a few paragraphs, don't blame AI. It's not AI.

ah? i hope that's a hoke, but clearly his acc is mostly ai, not just these.

No hoke. His posts are well known to be written by him. Going back years.