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
Thanks for the offer, but I’d just run YABS and even if I did further testing whatever result that comes out would now be seen to be biased or manipulated so there is nothing to be gained.
First, keep in mind that those were just 2 tests, 1 with fio, one with my benchmark. That's why I stress so much that a single or a few runs aren't worth much. After all, one doesn't want to know how fast a VPS is at some given point in time but rather it's performance generally, during the day, during the night, with little load, with high load, etc, etc.
As for calling a VPS poor or great, frankly, I don't care a lot about read speeds; those are almost always more or less OK or even great. But I still run those read tests because for some user and use cases they are important (or felt to be important). My interest in read test is largely what other call, often justifiably, "abnormal", because those weird numbers often tell me a lot about the system.
Also keep in mind that the numbers for the Contabo NVMe were average numbers over more than 100 runs. There were certainly also run where Rd. Seq. was way north of 4000.
I called the above dedicatserver VPS disk performance poor because it is. A value of about 50 MB/s (ca 35 in one, ca 65 in the other) write tests is good for a rusty spindle, OK for an SSD and simply poor for an NVMe. For an NVMe I expect 100 MB/s with an OK one and about 150+ MB/s for a good one, when testing with 1 MB blocks/slices. But again, that was just 1 run and maybe at a very bad moment.
Well, someone is tired. His reply goes to the south and north.. While his believe goes to the west.
Yes and no. One (1) run did achieve that number. That doesn't mean much, and frankly, even if that were an average over 100 runs I wouldn't care a lot because read results by their very nature tend to be heavily cached, first by the VM then by the node and possibly also by the controller.
And yes, I'm sure that it really did achieve that number because one can't do a whole lot wrong in a test that simply, e.g. 2048 times reads 1 MB from a drive. What tells me more is that the sequential and random read results are different by roughly a factor of 10. That very strongly hints at caching, because caching sequential reads is very easy and can be done with a relatively small cache but caching random reads is harder and requires more memory and more complex algorithms.
Funnily I also sometimes see the inverse which looks weird enough that I spent quite some time probing deeper in my home lab on a server set up in different ways. The reason seems to be that multiple caches are at work with a VM, usually at least 2 and those caches can also be quite different (different weaknesses and strengths) if node and VM run different OSs. But to come to a final conclusion I'd need to do more testing for which I currently don't have the time.
Thank you for your response and explanation - it is appreciated.
I did not diss the fio author. I stated some (relatively mild) criticism on yabs which uses fio but my criticism was not about fio.
Besides, I 'dissed' neither of the two but OK, I can understand that some might have taken it that way.
Yet-Another-Bench-Script
v2021-05-28
https://github.com/masonr/yet-another-bench-script
## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ##
>Basic System Information:
Processor : Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
CPU cores : 2 @ 2599.998 MHz
AES-NI : ✔ Enabled
VM-x/AMD-V : ❌ Disabled
RAM : 3.9 GiB
Swap : 0.0 KiB
Disk : 79.1 GiB
>fio Disk Speed Tests (Mixed R/W 50/50):
>iperf3 Network Speed Tests (IPv4):
Provider | Location (Link) | Send Speed | Recv Speed
| | |
Clouvider | London, UK (10G) | 923 Mbits/sec | 924 Mbits/sec
Online.net | Paris, FR (10G) | 936 Mbits/sec | 925 Mbits/sec
WorldStream | The Netherlands (10G) | 919 Mbits/sec | 922 Mbits/sec
Biznet | Jakarta, Indonesia (1G) | 717 Mbits/sec | 53.7 Mbits/sec
Clouvider | NYC, NY, US (10G) | 864 Mbits/sec | 511 Mbits/sec
Velocity Online | Tallahassee, FL, US (10G) | 462 Mbits/sec | 494 Mbits/sec
Clouvider | Los Angeles, CA, US (10G) | 752 Mbits/sec | 462 Mbits/sec
Iveloz Telecom | Sao Paulo, BR (2G) | 676 Mbits/sec | 301 Mbits/sec
>Geekbench 5 Benchmark Test:
Test | Value
|
Single Core | 622
Multi Core | 1214
Full Test | https://browser.geekbench.com/v5/cpu/8199674
Maybe on a node, preferably an empty one, but not on a VPS. fio and my benchmark don't both lie.
Your processor was OK and your network even quite good but your disks - on a VPS you sold - are poor. Says vpsbench and says fio.
What's next? Will you purchase a fat Ryzen or high end Xeon, run yabs on it and tell us that the results come from a VPS? Ridiculous.
You have a good basis (DC, network). Just buy and connect decent NVMes directly, don't overcrowd the node and you'll get decent results without publishing them yourself (which anyone nobody with a brain trusts).
if you agree we made the tests on same VPS and than put them here, on a full server node.
What do you say we do? me with CentOS , you with what do you want
Why, what for? I already did test a VPS from you (which I purchased) and I even ran fio - with exactly the yabs parameters (except 'posixaio instead of 'libaio') on it.
You could only lose. Either by achieving results which are similar to what I tested or by being confronted with the question why you sell VPSs with poor disk performance, but have other good VPSs in your back hand for special benchmarks.
This is not meant to attack or ridicule you. It's simply how I see it. Frankly, even running a benchmark yourself and publishing it yourself wasn't exactly smart, especially considering that the (IMO highly likely true) allegations against you (logging into customer VPS) tainted your credibility a lot.
Seriously, I think it would be better to create a new product with better drive performance and then have me and others benchmark it honestly.
take a video of u running the test with a spoon on your nose so we know its real
the YABS is made from another LET member after your statements
in another discussion
https://www.lowendtalk.com/discussion/173047/new-kvm-cloud-vps-2-core-4gb-ram-30gb-nvme-50gb-ssd-3-month-in-europe-eu#latest
You're not biased are you? This isn't because your servers suck, especially after the test and you just want some pay back?
Lol. I'm curious, hypothetically, how would such a post go that was in "good faith" in your opinion? If you saw my posts over the years on this in comparison, you'd think this was the greatest of faith by a million miles.
This is the speech that drunk surgeons give their nurse after the nurse told the doctor he can't operate today because he's too drunk.
First of all, thank you to @stevewatson301! this is a very well written OP and you've managed to stay civil throughout as I never could do.
To me, running vpsbench gives me no meaningful number and is just nonsensical to me. This is obvious in running it just once. I know I can copy a large file and reproduce Yabs disk results, but I can't fathom any real world test that could even come to close to replicating the numbers vpsbench gives. It's just useless.
Give a real world test that is in the same ballpark as vpsbench and the app will be validated. But giving numbers higher than RAM speeds, higher than SATA protocol, higher than fixed I/O limits, is just ridiculous.
What other alt accounts do you have here?
I predict this to be the longest war the history of humanity.
Even if I assume positive intentions from you (which I reasonably can't, see the opening post for why this is the case), you could always be given empty nodes which would produce a skewed benchmark, even if I use YABS.
Unfortunately, even after all this time, you haven't addressed my actual question and only got a bunch of BS speculation and unwarranted namecalling ("can you code" etc. without any factual proof that I was incapable of such). I'm taking for what it is, the benchmark is still inaccurate and you have no proper answer for it.
I don't have any, don't need to hide behind them like apparently you do. An endorsement by you to the OP doesn't give him more rep, it lowers it because of the weight of yours, which is heavy and of low quality.
in timbos defense he equally hates everyone
Yeah... I know
One, who bare "king" in his title will be challenged and scrutinized by the public. Title is not only tag,but also mark of authority in specific field.
Those who not sucumb to bashing and questioning - will prevail.
@jsg endure and you will earn some respect here. Fall to a trap of pride and cockines - you will be defamed and mocked.
Look like you are old enough to read, please read a lot on LET especially discussion within 3 or 4 months before on Chest Pit, Cociu, Dedicat, and Contabo threads before made such clueless & pointless comment.
Be like Robert de Niro, he is good performer and know how to be a good guy.
I've been here longer than you have, so I don't understand your POV. I've read all those threads, what's that got to do with this one?
Stop eating too much chocolate, it's messing with your brain
Nah your account dead a long time ago. And recently you just got ressurcted (July 2021) and post some comments. I suggest you read all the discussion I mentioned above when your account were dead. Since you are clueless old man.
I was lurking, what does that have to do with this???
You really are an idiot.
Nah you are definitely clueless old man. Hahaa.. You are better lurking than post a clueless & pointless comment. Sorry @stevewatson301 I derailed your thread a bit.
Even when other members, including reputed providers, pointed out that Hostsolutions was selling off hardware and IPs, jsg repeatedly came up with wrong explanation of how hardware and BGP announcements worked and when that was proven wrong too, he resorted to the kinds of "you're clueless", "your version of truth vs. my version of truth" etc. arguments that you see here.
You can read that thread for more info. Take @chocolateshirt's advice, he really is right in this case