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.
Comments
The $7/year is a common expression on this forum symbolizing a very lowend, cheap vps, it is not meant to be taken literally.
Hi rcy026
You are missing the point by focusing on the price tag.
Whether the VPS costs $7, $15, or $50, it does not change the flow of the technical argument.
You could be paying $50/month for a plan with High CPU and massive RAM. But having high compute resources does not magically fix the I/O bottleneck.
*Why? Because the speed of reading and writing data is strictly bound by the IOPS/Throughput limit (QoS) set in the hypervisor config. If the provider hard-codes a limit, all that expensive RAM and CPU sit idle while waiting for the disk.
We are not asking for 'Dedicated Performance' for $7. We are saying: If you sell 'High Performance NVMe', don't deliver a software-choked HDD performance, regardless of the price.
As people have tried to explain to you repeatedly in your other thread about this, the limit is not hard coded. A lot of users have shown benchmarks that clearly exceeds 200Mb/s.
The reason you hit the limit is most likely that you abused the disks which affected every other user, so your throughput is limited. This behavior is industry standard and has been basically since they invented virtualization.
You seem to think that people does not understand the technical facts or what the limits mean. Let me remind you that this forum is populated by providers and people that work with these kind of technologies every day. We totally understand what you are saying and what Ionos is doing, we simply do not agree with your opinion. Not agreeing is not the same as not understanding.
I do agree with you that technical transparency is good though, absolutely, limits like this should be mentioned. On the other hand, I've been in the hosting business for 40 years, and it has always been like this, it's just the way it works and everybody knows that.
And to be honest, expecting something else is just naive, borderline stupid. It's like buying a car and then complain at the dealer that you can not consistently travel at the cars top speed because of traffic, and demand that the dealer removes every other car from the road. It does not say when you buy the car that you will not be able to always travel at top speed, but still, expecting it is just unreasonable and everyone understands that.
Given this is claimed to be a technical thread rather than a PMS thread, can you give sample QEMU flags on how to impose these two limits?
We'd like to experiment with this QoS limits feature and how to detect the QoS limits through machine learning techniques.
Flagged this shit so fast like you wouldn't believe
If this is still about IONOS, you do need to remember that they are priced extremely low compared to other big operators, there is something to be said about getting what you pay for. In europe, their lowest tier VPS plans are just 1-2 euro per month with 10GB or 80GB of storage. At that price, most of it is going towards the IPv4 in my opinion, let alone if you contact their support team at all.
What are you running that you need more than a few hundred megabytes of sustained speed? Even 125MB/s of disk speed is enough to fully saturate a gigabit uplink, and databases are more IOPs driven than raw speed driven, does IONOS cap IOPs the same way?
I get your point and it is some level of mis-advertisement, but overselling on resources isn't new for these ultra low budget VPS hosts, it happens with CPU, Network and in extreme cases, especially on summerhosts, on real RAM.
I doubt you need to use ML to detect it. It's not exactly designed to be stealthy, just using traditional leaky bucket throttling.
4000 IOPS throttle, 10 second 9000 IOPS burst, 200 MB/s throughput, 20 second 900 MB/s burst:
You can also throttle read and write separately, e.g.
bps_wr=200000000,bps_rd=400000000to have 200 MB/s write but 400 MB/s read (and you can do the same with IOPS). And you can crate throttling groups to apply to multiple disks at once.More info: https://github.com/qemu/qemu/blob/master/docs/throttle.txt
Proxmox configuration is equally straightforward, but with slightly different parameters (e.g.
mbps_wr). That's how most hosts will be configuring it, anyway. Real-life example (VPS in Brazil from JustHost):So I've got a 700 MB/s throttle for both read and write on that machine with no artificial IOPS limit and no burst on that VPS. Host pagecache is not used, TRIM commands are passed through, and it's 21 GB (20 GiB) in size, announcing itself to the guest as an SSD. Good enough for me.
in Vietnam, manufacturers often print a shrimp on instant noodle packages. Maybe I should sue them too for false advertising, right?
when we use services from cheap providers, by default we also have to accept the idea of “you get what you pay for.” As long as it is good within its price range, that’s fine. We can’t demand performance beyond its price level, then, when it doesn’t meet our expectations, keep writing one complaint after another against them.
An SSD with a speed of 200 MB/s is still many times faster than an HDD in real-world VPS usage conditions, so how can you say the speeds are the same?
The discussion from "The “NVMe” Illusion — Why Your “High Performance” VPS Feels Like a Spinning Disk" has been merged into this thread.
I can't say anything about other hosters, but when we indicate that we are using SSD or NVMe, we simply indicate which type of storage is being used. Just like when we talk about the type of processor: Intel or AMD. User expectations don't matter to us. Synthetic tests of the results can help the customer make the right choice according to his requests.
Hi,
just to give another idea @Tufan
You wrote you bought XL ( 480 GB ) from Ionos.
Usually the physical disk size used in such environments providing storage is 6.4 TB or bigger.
So with your 480 GB you bought 1/13th of the "space" capacity of such a drive. Following your own logic so far that should also mean that you bought 1/13th of "speed" capacity of such a drive.
A normal NVMe u.2/u.3 drive that we could expect in such an environment of a hoster will do maybe 5GB/s in raw speed ( if ever -- probably less ). So you bought ~ 5 GB / 13 = 384 MB/s
And thats actually without any redundancy. So if we consider a N+2 redundancy ( hopefully the minimum that ionos will run ) the real portion of performance you bought with your 480 GB disk would be somewhere around 200-250 MB/s.
To me it basically looks that you get what you paid for essentially, following your own logic about providers should not oversell and so on. In that logic customers should not overuse. Right? So the 10 GB XS Ionos server should be only allowed to use 1/640th of a drive speed. As he actually just paid for 1/640th of the drive capacity. -- Luckily for the 10 GB customer this is not the case ( hopefully ).
Dont get me wrong, i am fully with you complaining about providers who limit what ever but do not communicate it. Thats in my personal opinion hard customer fooling.
But on the other hand, also customers have some responsibility to understand technically what they actually buy there and for how much portion of performance they actually pay and have a right on.
Looking at it from another positive angle: if IONOS didn’t impose such soft limits, then on a server with maybe a dozen LET members buying VPSs there, with the habit of running YABS nonstop—some guys even write bash scripts to run YABS occasionally so their VPS doesn’t feel “idle”—the remaining users probably wouldn’t be able to use their VPS at all.
So think positively. Life isn’t that long anyway; thinking negatively is bad for your heart, liver, and kidneys 🙂
Why was the other thread about NVME software throttling, which had numerous posts, deleted?
It wasn't, it was merged into this one.
@Tufan 🤤🤤
https://www.robovps.biz/en/hi_cpu_vps_in_finland/372-antares.html
Tue Jan 13 07:55:25 PM MSK 2026
Basic System Information:
Uptime : 0 days, 0 hours, 17 minutes
Processor : AMD Ryzen 9 9950X 16-Core Processor
CPU cores : 8 @ 4291.920 MHz
AES-NI : ✔ Enabled
VM-x/AMD-V : ✔ Enabled
RAM : 7.8 GiB
Swap : 0.0 KiB
Disk : 177.1 GiB
Distro : Debian GNU/Linux 13 (trixie)
Kernel : 6.12.57+deb13-amd64
VM Type : KVM
IPv4/IPv6 : ✔ Online / ✔ Online
ISP : Scalaxy B.V.
ASN : AS58061 Scalaxy B.V.
Host : Gtelcom LLC
Location : Helsinki, Uusimaa (18)
Country : Finland
fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/vda2):
iperf3 Network Speed Tests (IPv4):
iperf3 Network Speed Tests (IPv6):
Geekbench 6 Benchmark Test:
Test | Value
|
Single Core | 3350
Multi Core | 14271
Full Test | https://browser.geekbench.com/v6/cpu/16069021
YABS completed in 11 min 2 sec
Just my $0.02 on this topic...
For full disclosure, years ago I was a VERY small player in the hosting business but today am much more of a customer than a provider and manage something like two dozen VPS's from about 18 different providers, and another two rather large dedicated servers.
What @Tufan needs to realize is that his VPS doesn't have any NVME disk drives. He has virtual disks. It's the hardware that has NVME drives and, in my opinion, it's perfectly acceptable for Ionos to advertise the type of hardware they use to deliver their VPS services as long as they aren't making false claims about the expected performance.
I think @Tufan equates seeing NVME on the hardware to receiving full NVME performance on his virtual disks which is just not going to happen on a shared system.
I truly understand the frustration of expectations not being met, but I don't place blame on Ionos other than their inability to adequately communicate the situation.
This is a really good point.
I just want to know what kind of HDD VPS @Tufan has that they hit 200MB/s on 4k Random read (or write). I will buy.
the issue was sequential
And it's not only the storage devices that have to be thought of in this way. Even if you have 0% CPU steal, there's still going to be hypervisor overhead, and you may find that you can't push a single core about 80% in some situations. And even when running a workload with few VMExits, you're virtually always going to be passed a cold cache for your timeslice.
And then your "1 Gbps port" is going to be shared as well. Even if it doesn't say "shared", VirtIO still has overhead, and most small hosts do not use pass through their NIC with SR-IOV.
Really the only things that are "yours" is the total storage capacity you're given and your RAM capacity (but not speed). Even with RAM, they might be using memory ballooning (although you can turn that off).
It's impossible to take a virtual machine and directly compare its performance to a physical machine.
That sounds right for a VDS, you get consistent performance your neighbor doesn't impact you so much.
In shared, you hope to exceed average performance when not fully busy to offset the lower performance when busy. Otherwise, you're limiting the duty cycle and efficiency and limiting overall performance.
Always having lower performance regardless of actual usage sucks.
Cool story, bro.
Really? I thought it was obtuse. The provider intentionally advertises NVMe but you shouldn't expect certain performance? That's silly. He'd have a point if they didn't mention NVMe at all.
If they said they used Dell or HP branded servers, no expectation of performance by just that.
You're entitled to believe that just as Tufan is entitled to feel cheated but unfortunately, that's just not the way the world works. When you purchase a VPS you're purchasing a service. If the provider delivers the service they promised, you got what you paid for. If you want a performance guarantee higher than the standard service offering, you typically need to purchase a SLA (if offered).
He's not expecting a performance guarantee, nor was he expecting artificially limited performance he was unaware of. He was expecting better performance than mechanical drives.
If you were offered standard service and premium service, you'd expect premium to have better performance than standard, without needing any specific guarantee, no? And then be told, "well, it's premium hardware to make our lives better but it's no more premium than standard for you."
Since I'm really not into the online debate thing and I've already put in more than the $0.02 worth I started with on my initial post, this will be my last post on this subject. So, whatever your response, you win!
Regarding his expecting better performance than mechanical, like I said previously, I understand the frustration when expectations aren't met. If he wants that better performance guaranteed to him, then it sounds an awful lot like he WAS expecting a performance guarantee.
Regarding my expectations on premium vs. standard services, I would expect some providers' standard service will outperform other providers' premium service and if I required some level of guaranteed performance with either, I'd look at the SLA.
~Namaste, Brother!~
R/W 500+500 means that the disk is capable of doing 1000+ MB/s read or write when used alone. A 50/50 mix is a much harsher job. Maybe you don't even understand how benchmark results are made.
The main reason people avoid HDDs is their diabolical IOPS. But this VPS seems not to have that issue. You didn't pay for the entire disk's price, so why would you expect to get full performance? Yeah limiting speed from the start is bad. I won't use IONOS for this reason, but it's the wrong way to complain, imo.
lets get technical for a second:
disk iops are shared, not dedicated
The implication is clearly higher performance but if you care about the minimums maybe you ask more pointed pre-sale questions. Perhaps I should have said it's a good way to look at it.