Overselling, and when the customer discovers the issue, the response is like this =))

edmarhub

@rpqu said:
What happened to @Snitchboss123, @angstrom ?

I think because of this thread, he was ragebaiting 😜

https://lowendtalk.com/discussion/comment/4870381#Comment_4870381

zed

he offered refund, what else do you want?

forest

@zed said:
he offered refund, what else do you want?

An apology, of course!

x0x0x

As others have mentioned, the benchmark results still look usable depending on what you're using it for. At the end of the day, it's still a VPS. Resources are shared so performance can and will vary depending on what other users are doing at the time, and occasionally you are going to get noisy neighbors.

mhpteam

guys he's Vietnamese, and how yall accept to pay for 8 cores 9950X just to get 1100 multi core marks 💀

rcy026

Is it just me, or was that not a very honest and to the point answer from the provider?

They simply state facts and tell it like it is, they do not try to cover it up or come with excuses. They admit the node is loaded and they can not fix it right now and therefor they offer a refund. I would without hesitation accept that answer and live with half-decent performance until they get the opportunity to upgrade. After all, this is a vps, it is all shared resources so half-decent performance is what you should expect.

I do not have any vps with @DigitalFyre but this thread makes me want to get some, an honest provider is hard to find these days. This was probably not what op intended but he just gave them a new customer.

zed

@rcy026 said: Is it just me, or was that not a very honest and to the point answer from the provider?

yea i mean my eyes glaze at the ai phrasing and i get the op being annoyed at the situation but seemed like it was handled more than appropriately. different folks i guess.

tiendungdev

@DigitalFyre said:
First, I apologize to everyone reading this for the lengthy reply.

Hi @tiendungdev !

You're welcome to tag us in the post if you have an issue, and if the ticket response is unsatisfactory, please state that in the same ticket.

Your VPS is in Singapore, and it's been a sold-out location for a while, yet we continue to receive deployment requests, all of which we decline.

None of our nodes are oversold. We publicly shared the formula through which we determine how many VMs are hosted on a node:

Node Total RAM minus 8 GB RAM = Total RAM available for VM deployments.

Example: 192 - 8 = 184 GB available for VMs, and we fill nodes based on that.

As the ticket states, Singapore is currently pretty full, and there is no other node to move you to; we also offered a refund for your last payment if the performance does not meet your standards. We also state publicly that performance fluctuates as nodes fill up.

Is this performance ideal? Absolutely not. Do we want to do something about it? Yes, we do. However, the hardware market is horrible, and Singapore has been a hot location. I told other customers this, and John did the same. We want to improve things, but our hands are tied, and the only thing we can do right now is either:

1. Throttle everyone's disk use on the node, which is a bit restrictive and overkill, but we might have to, and if we do, we'll send notice to all customers hosted on the node.
2. Offer refunds for unsatisfied customers.

For what it's worth, we are offering these refunds outside our refund policy, and we process them immediately if the customer accepts, without any delay whatsoever. If the response comes in during off hours, we process the refund first thing the next morning. We've even refunded "Funds" which customers added to their accounts. Why? Because I don't want someone's money when they are not happy with the service they received, and we don't want to keep the $10, $20, or even $100 paid for the service if the customer is suddenly unhappy. Even if you paid a year in advance, we'd refund your remaining balance, including the month you became unhappy with the service. This is why we don't offer annual billing for VPS plans.

We cannot and will not claim things are honky dory when they aren't, and we have never done so when we received complaints. We always take action and do what we can.

From this post, I understand you're unhappy with the performance and seem indifferent to the refund offer. To be completely honest, I'm not entirely sure how we can make this right.

Furthermore, we never claimed we had a solution in place, and we have been fully transparent about the situation.

@MikeA said: AI generated reply I'd immediately start moving stuff off either way lol.

The response is not fully AI-generated. We use an AI Assistant tool that can reformulate a response or generate one based on the conversation and past cases; then the person posting the response edits it as they see fit.

We are also testing automated AI-agent ticket responses, which are signed by the AI agent and clearly state that an "AI Assistant" is posting the response.

Prometheus - AI Support Agent, DigitalFyre

The staff and I also review those responses, and if something is inaccurate, we correct it in the next response and own the mistake.

Some of these responses are predefined because we receive the same inquiry repeatedly, and it makes little sense to write the same response by hand over and over.

We also use Grammarly to check response tone and help ensure responses aren't accidentally passive-aggressive or unprofessional. In fact, I ran this response through Grammarly both as I wrote it and after, and I am not hiding that.

If anyone has a question or needs additional clarity, feel free to tag me here in this post, or on Discord.

@DigitalFyre said:
First, I apologize to everyone reading this for the lengthy reply.

Hi @tiendungdev !

You're welcome to tag us in the post if you have an issue, and if the ticket response is unsatisfactory, please state that in the same ticket.

Your VPS is in Singapore, and it's been a sold-out location for a while, yet we continue to receive deployment requests, all of which we decline.

None of our nodes are oversold. We publicly shared the formula through which we determine how many VMs are hosted on a node:

Node Total RAM minus 8 GB RAM = Total RAM available for VM deployments.

Example: 192 - 8 = 184 GB available for VMs, and we fill nodes based on that.

As the ticket states, Singapore is currently pretty full, and there is no other node to move you to; we also offered a refund for your last payment if the performance does not meet your standards. We also state publicly that performance fluctuates as nodes fill up.

Is this performance ideal? Absolutely not. Do we want to do something about it? Yes, we do. However, the hardware market is horrible, and Singapore has been a hot location. I told other customers this, and John did the same. We want to improve things, but our hands are tied, and the only thing we can do right now is either:

1. Throttle everyone's disk use on the node, which is a bit restrictive and overkill, but we might have to, and if we do, we'll send notice to all customers hosted on the node.
2. Offer refunds for unsatisfied customers.

For what it's worth, we are offering these refunds outside our refund policy, and we process them immediately if the customer accepts, without any delay whatsoever. If the response comes in during off hours, we process the refund first thing the next morning. We've even refunded "Funds" which customers added to their accounts. Why? Because I don't want someone's money when they are not happy with the service they received, and we don't want to keep the $10, $20, or even $100 paid for the service if the customer is suddenly unhappy. Even if you paid a year in advance, we'd refund your remaining balance, including the month you became unhappy with the service. This is why we don't offer annual billing for VPS plans.

We cannot and will not claim things are honky dory when they aren't, and we have never done so when we received complaints. We always take action and do what we can.

From this post, I understand you're unhappy with the performance and seem indifferent to the refund offer. To be completely honest, I'm not entirely sure how we can make this right.

Furthermore, we never claimed we had a solution in place, and we have been fully transparent about the situation.

@MikeA said: AI generated reply I'd immediately start moving stuff off either way lol.

The response is not fully AI-generated. We use an AI Assistant tool that can reformulate a response or generate one based on the conversation and past cases; then the person posting the response edits it as they see fit.

We are also testing automated AI-agent ticket responses, which are signed by the AI agent and clearly state that an "AI Assistant" is posting the response.

Prometheus - AI Support Agent, DigitalFyre

The staff and I also review those responses, and if something is inaccurate, we correct it in the next response and own the mistake.

Some of these responses are predefined because we receive the same inquiry repeatedly, and it makes little sense to write the same response by hand over and over.

We also use Grammarly to check response tone and help ensure responses aren't accidentally passive-aggressive or unprofessional. In fact, I ran this response through Grammarly both as I wrote it and after, and I am not hiding that.

If anyone has a question or needs additional clarity, feel free to tag me here in this post, or on Discord.

I appreciate the honesty shown in your ticket acknowledging the issue, but the practice of overselling to the point of overloading raises serious questions about business ethics?!. While I understand that performance might drop compared to when I first purchased the service (when the server had fewer clients), the current level of degradation is unacceptable.

Naturally, I will be moving my business elsewhere, as there are plenty of good providers out there.

Please note that I am not a new customer—I have been with you since the initial launch—so this kind of profit-driven overcrowding is particularly disappointing.

tiendungdev

@rcy026 said:
Is it just me, or was that not a very honest and to the point answer from the provider?

They simply state facts and tell it like it is, they do not try to cover it up or come with excuses. They admit the node is loaded and they can not fix it right now and therefor they offer a refund. I would without hesitation accept that answer and live with half-decent performance until they get the opportunity to upgrade. After all, this is a vps, it is all shared resources so half-decent performance is what you should expect.

I do not have any vps with @DigitalFyre but this thread makes me want to get some, an honest provider is hard to find these days. This was probably not what op intended but he just gave them a new customer.

Ok, good luck

alincupunct

@rcy026 said:
Is it just me, or was that not a very honest and to the point answer from the provider?

They simply state facts and tell it like it is, they do not try to cover it up or come with excuses. They admit the node is loaded and they can not fix it right now and therefor they offer a refund. I would without hesitation accept that answer and live with half-decent performance until they get the opportunity to upgrade. After all, this is a vps, it is all shared resources so half-decent performance is what you should expect.

I do not have any vps with @DigitalFyre but this thread makes me want to get some, an honest provider is hard to find these days. This was probably not what op intended but he just gave them a new customer.

Great idea when they can't be bothered to respond to tickets and just use AI )

rcy026

@alincupunct said:

@rcy026 said:
Is it just me, or was that not a very honest and to the point answer from the provider?

They simply state facts and tell it like it is, they do not try to cover it up or come with excuses. They admit the node is loaded and they can not fix it right now and therefor they offer a refund. I would without hesitation accept that answer and live with half-decent performance until they get the opportunity to upgrade. After all, this is a vps, it is all shared resources so half-decent performance is what you should expect.

I do not have any vps with @DigitalFyre but this thread makes me want to get some, an honest provider is hard to find these days. This was probably not what op intended but he just gave them a new customer.

Great idea when they can't be bothered to respond to tickets and just use AI )

Well, if the response is correct and answers my question I do not care who writes it.
And they claim to check each answer even if it's written by AI, so saying that they "can't be bothered to respond to tickets" is not really a fair judgement.

Customer service is one of those areas where I do not mind correctly implemented AI, since the average AI is usually more intelligent then the average customer. Not trying to offend anyone, but after 30 years in IT my expectation on the average customer is unfortunately extremely low not to say non-existent.

kenjing789

@mhpteam said:
guys he's Vietnamese, and how yall accept to pay for 8 cores 9950X just to get 1100 multi core marks 💀

With price 10$ for 8c8g. I dont think you can expect extreme fast 8c with 8g ram at all on a 16/32 cpu. Ram is expensive u know

mhpteam

@kenjing789 said:

@mhpteam said:
guys he's Vietnamese, and how yall accept to pay for 8 cores 9950X just to get 1100 multi core marks 💀

With price 10$ for 8c8g. I dont think you can expect extreme fast 8c with 8g ram at all on a 16/32 cpu. Ram is expensive u know

okay nvm, the commitment must be 4:1 to compensate this 💀

alincupunct

@rcy026 said:

@alincupunct said:

@rcy026 said:
Is it just me, or was that not a very honest and to the point answer from the provider?

They simply state facts and tell it like it is, they do not try to cover it up or come with excuses. They admit the node is loaded and they can not fix it right now and therefor they offer a refund. I would without hesitation accept that answer and live with half-decent performance until they get the opportunity to upgrade. After all, this is a vps, it is all shared resources so half-decent performance is what you should expect.

I do not have any vps with @DigitalFyre but this thread makes me want to get some, an honest provider is hard to find these days. This was probably not what op intended but he just gave them a new customer.

Great idea when they can't be bothered to respond to tickets and just use AI )

Well, if the response is correct and answers my question I do not care who writes it.
And they claim to check each answer even if it's written by AI, so saying that they "can't be bothered to respond to tickets" is not really a fair judgement.

Customer service is one of those areas where I do not mind correctly implemented AI, since the average AI is usually more intelligent then the average customer. Not trying to offend anyone, but after 30 years in IT my expectation on the average customer is unfortunately extremely low not to say non-existent.

Enjoy your 8 vcore 9950x getting 1.1k in gb6 then, buy more from dogshit providers like this.

DigitalFyre

@tiendungdev said: I appreciate the honesty shown in your ticket acknowledging the issue, but the practice of overselling to the point of overloading raises serious questions about business ethics?!. While I understand that performance might drop compared to when I first purchased the service (when the server had fewer clients), the current level of degradation is unacceptable.

The unacceptable degradation is completely understandable, and we acknowledged it and offered a refund. As stated, right now, there's not much we can do except see what we can do about reasonable global throttling on disk and CPU use due to the number of VMs on the node. For what it's worth, this node has fewer than 60 VMs. Of course, even with cancellations, we will keep the location marked as OOS and won't deploy any additional services on that node.

Naturally, I will be moving my business elsewhere, as there are plenty of good providers out there.

I'm sorry to see you go, but it's also completely understandable. I refunded your most recent invoice (September, 2026), and you are not obligated to pay your outstanding invoice if you're moving your service to another provider. Please reach out to our team if you need additional time to migrate so I can disable automatic service suspension to help you migrate without any additional stress.

Please note that I am not a new customer—I have been with you since the initial launch—so this kind of profit-driven overcrowding is particularly disappointing.

That's true that you have been a customer since our first LET offer. Thank you for the time you've spent with us and for using our services.

I truly wish there were something more impactful I could do about Singapore, and we have been working hard to secure more hardware in the region. Sadly, our efforts have been unsuccessful so far.

DigitalFyre

@rcy026 said: Customer service is one of those areas where I do not mind correctly implemented AI

Plenty of repeated questions, although answered in the Knowledge Base, are still being asked. This is where automated responses from the AI assistant come in handy.

Also, I'd like to reiterate that the AI responses are still being reviewed by support staff, more often than not, by me. If there's a mistake or inaccuracy, we correct it and communicate the correction to the customer.

I support using AI to a certain degree. However, I'm not comfortable relying on it fully. We even have a banner in the AI Assistant chat that states, "AI can make mistakes. Please check important information.".

DigitalFyre

@kenjing789 said: With price 10$ for 8c8g. I dont think you can expect extreme fast 8c with 8g ram at all on a 16/32 cpu. Ram is expensive u know

Last I checked, 48gb ECC is around ~$2k per stick.

DigitalFyre

@zed said: he offered refund, what else do you want?

@forest said: An apology, of course!

OP received both

DigitalFyre

@forest said: How about doing some profiling to see what the typical use is and use throttling with a generous burst limit?

Literally looking into this right now

rpqu

@DigitalFyre , sorry to ask this. But can you tell me the drives inside that hypervisor?
Assume usable space is 1800GB. His share is 60GB. Ratio, 30:1
While FIO at 4k it's 260.K:18.6K. So, 14:1
Perhaps the disks are capable of delivering more than 558K iops at 4k?

DigitalFyre

@rpqu said: @DigitalFyre , sorry to ask this. But can you tell me the drives inside that hypervisor?

In Singapore, they are Nytro 5550M SSD 3.2TB XP3200LE70035 Gen4 NVMe. I have a "spare" disk on the node, so I could move some VMs to it. It might help with IO, but honestly, we need more hardware, and the first thing is to offload a good chunk of users from the old HV to the new one instead of going with some bandaid solution 🤔

rpqu

@DigitalFyre said:

@rpqu said: @DigitalFyre , sorry to ask this. But can you tell me the drives inside that hypervisor?

In Singapore, they are Nytro 5550M SSD 3.2TB XP3200LE70035 Gen4 NVMe. I have a "spare" disk on the node, so I could move some VMs to it. It might help with IO, but honestly, we need more hardware, and the first thing is to offload a good chunk of users from the old HV to the new one instead of going with some bandaid solution 🤔

Hmmm

Seagate Nytro 5550M 3.2 TB — XP3200LE70035

Spec	Value
Capacity	3.2 TB
Interface	PCIe Gen4 ×4, NVMe
Form factor	2.5" (15 mm; U.2/U.3-class)
NAND	3D eTLC
Endurance tier	Mixed use — 3 DWPD
Total writes (TBW)	~17,500 TB (≈17.5 PB over 5 yr)
Seq. read	7,400 MB/s
Seq. write	3,300 MB/s
Random read	1,150,000 IOPS (4 KB)
Random write	190,000 IOPS (4 KB)
MTBF	2.5 M hours
UBER	1 sector per 10^17 bits read
Security	Hardware AES self-encryption (TCG); SED variant
Power-loss protection	Yes (enterprise capacitor-backed)
Warranty	5-year limited

Say usable space is 3000GB -> 1:50
OP should have receive 190/50= 3.8K iops. But he received 9.8k iops.
Maybe you should sell the spare disk/capacity to OP for additional $50/month 😂

DigitalFyre

@rpqu said: Say usable space is 3000GB -> 1:50

OP should have receive 190/50= 3.8K iops. But he received 9.8k iops.
Maybe you should sell the spare disk/capacity to OP for additional $50/month 😂

😂 Nah! I wouldn't do that! Honestly, if you look at the node's load with a little less than 60 VMs, over the past 4 weeks, it's not unreasonable, and the RAM use is expected since we commit memory to users' VMs:

Node YABS:

fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/nvme0n1p3):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 1.31 GB/s   (320.6k) | 1.98 GB/s    (30.2k)
Write      | 1.31 GB/s   (321.4k) | 1.99 GB/s    (30.4k)
Total      | 2.63 GB/s   (642.1k) | 3.98 GB/s    (60.7k)
           |                      |                     
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 1.58 GB/s     (3.0k) | 1.59 GB/s     (1.5k)
Write      | 1.67 GB/s     (3.1k) | 1.69 GB/s     (1.6k)
Total      | 3.26 GB/s     (6.2k) | 3.28 GB/s     (3.1k)

iperf3 Network Speed Tests (IPv4):
---------------------------------
Provider        | Location (Link)           | Send Speed      | Recv Speed      | Ping           
-----           | -----                     | ----            | ----            | ----           
Clouvider       | London, UK (10G)          | 1.14 Gbits/sec  | 1.46 Gbits/sec  | 150 ms         
Eranium         | Amsterdam, NL (100G)      | 6.13 Gbits/sec  | 7.15 Gbits/sec  | 160 ms         
Uztelecom       | Tashkent, UZ (10G)        | 1.64 Gbits/sec  | busy            | 192 ms         
Leaseweb        | Singapore, SG (10G)       | 8.95 Gbits/sec  | 9.08 Gbits/sec  | 1.77 ms        
Clouvider       | Los Angeles, CA, US (10G) | 625 Mbits/sec   | 1.33 Gbits/sec  | 166 ms         
Leaseweb        | NYC, NY, US (10G)         | 5.11 Gbits/sec  | 5.62 Gbits/sec  | 211 ms         
Edgoo           | Sao Paulo, BR (1G)        | busy            | 3.72 Gbits/sec  | 332 ms         

Geekbench 6 Benchmark Test:
---------------------------------
Test            | Value                         
                |                               
Single Core     |                               
Multi Core      |                               
Full Test       | https://browser.geekbench.com/v6/cpu/19260038

YABS completed in 9 min 0 sec 

FIO for the VM storage drive, not so great:

fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/nvme1n1p1):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 160.51 MB/s  (39.1k) | 184.57 MB/s   (2.8k)
Write      | 160.93 MB/s  (39.2k) | 185.54 MB/s   (2.8k)
Total      | 321.45 MB/s  (78.4k) | 370.12 MB/s   (5.6k)
           |                      |                     
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 183.64 MB/s    (350) | 183.99 MB/s    (175)
Write      | 193.40 MB/s    (368) | 196.25 MB/s    (187)
Total      | 377.04 MB/s    (718) | 380.25 MB/s    (362)

So OP definitely has a valid point regarding disk performance.

However, none of this matters as much as the fact that we still need to deploy additional hardware and lighten the load on this node.

What I can, and will, do is move some VMs over to the other drive and monitor the performance.

host_c

@WyvernCo said:
The stats for $3/mo is fine BUT they are being advertised as "High-performance AMD Ryzen & EPYC KVM VPS with blazing NVMe storage" The speeds are fine for many use cases but they are not NVMe speeds. Provider has oversold versus what they advertised.

3$/Mo and some are pissed it is a bit oversold/cramed?

@rpqu said: Say usable space is 3000GB -> 1:50
OP should have receive 190/50= 3.8K iops. But he received 9.8k iops.
Maybe you should sell the spare disk/capacity to OP for additional $50/month 😂

dafq @rpqu , so normal capped speeds for NVME should be 3.8K IOPS, we have that on HDD, I think we need to revise the storage specs

DigitalFyre

OK! I tested this out and moved a VM from the crowded disk to the other one. Things are most certainly significantly better.

fio Disk Speed Tests (Mixed R/W 50/50) (Partition /dev/sda3):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 481.94 MB/s (117.6k) | 2.07 GB/s    (31.6k)
Write      | 483.21 MB/s (117.9k) | 2.08 GB/s    (31.7k)
Total      | 965.16 MB/s (235.6k) | 4.15 GB/s    (63.3k)
           |                      |                     
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 1.72 GB/s     (3.2k) | 1.58 GB/s     (1.5k)
Write      | 1.81 GB/s     (3.4k) | 1.68 GB/s     (1.6k)
Total      | 3.53 GB/s     (6.7k) | 3.26 GB/s     (3.1k)

With that said, I'll reach out to the customers who had disk performance issues and offer to move their VMs to help them get better performance. Also, I think I may have located a couple of I/O Hogs, whom I'll reach out to after throttling is applied.

Again, this is not a permanent fix. It's something to keep the node in a decent state until we procure more hardware.

rpqu

@host_c said:

@rpqu said: Say usable space is 3000GB -> 1:50
OP should have receive 190/50= 3.8K iops. But he received 9.8k iops.
Maybe you should sell the spare disk/capacity to OP for additional $50/month 😂

dafq @rpqu , so normal capped speeds for NVME should be 3.8K IOPS, we have that on HDD, I think we need to revise the storage specs

On paper, yes. The guaranteed IO is 3.8K IOPS at 4k per 60GB on that drive. Is my calculation wrong?

zed

@DigitalFyre said: Again, this is not a permanent fix. It's something to keep the node in a decent state until we procure more hardware.

not a customer but i appreciate your attitude, very classy.

DigitalFyre

@zed said: not a customer

...Yet

but i appreciate your attitude, very classy.

Thank you very much!

I would also like to thank everyone else in the thread who kept the discussion constructive and classy, and refrained from jumping to conclusions and using unsavory language.

---

I am still working on that node to figure out what really is causing the slowness.

I found conclusive data showing the disk is perfectly healthy and has virtually no wear. However, after writing a very large amount of data (887 TB) over its life, the drive's internal housekeeping appears to have fallen behind. ( I don't know how else to describe it )

One user had 600 million write requests and 1.2 billion read requests within a 50-day span on a 20 GB disk.

That said, I will take a moment to admit that the initial diagnostics and the quick assumption that it's just because there are a lot of VMs don't live up to the standards we hold ourselves to.

I'll keep doing what I need to do and schedule a maintenance window to do some housekeeping.

deafcon

I'm real impressed with the way DigitalFyre has handled this thread.

rpqu

@DigitalFyre said:
One user had 600 million write requests and 1.2 billion read requests within a 50-day span on a 20 GB disk.

That person should be throttled lol. 12M write request per day, 500K per hour. 138iops. Wtf this guy doing