i guess @Not_Oles gonna one of these shortly!
@seenu said at https://talk.lowendspirit.com/discussion/comment/55523/#Comment_55523
i guess @Not_Oles gonna one of these shortly!
I would get one in a flash if a few people convinced me they would come aboard. Who's in?
Greetings from Mexico! ??
Comments
@Not_Oles What would be the pricing like? same as your exclusive advertisement or different?
@Asim
The AX101 server costs € 99.00 per month, plus additional IP addresses., for example € 26.89 monthly for a /27. Could add a 12 TB SATA for € 24.00.
Base specs are 16 cores, 32 threads, 128 GB ECC, and 2 x 3.84 TB NVMe.
What plans and prices would you like to see?
A big attraction of a box like that is using all the cores to compute something fast, and that doesn't really fit in with multiple users with vm's. I guess it's still quite a good cpu and ssd to cost ratio as a VM box though. Also I'd like to know the write endurance of the ssd's. I fear they might be QLC given the very low price.
Knowing Hetzner and the fact that they mention these as datacenter SSDs I'd assume that they have pretty decent endurance.
Irrespective, I also assume that if you're nearing end of life on these drives, they will replace them without issue (and quickly at that).
So all in all, I'd assume that you're covered as far as these drives go from a endurance standpoint.
I'll just tag @Hetzner_OL in case an official reassurance will help.
I must say that the recent Hetzner hardware line up refresh has been really great - the SX63 and now the AX101 - they both are really raising the bar on the performance-price ratio. If only more of us low-end-(l)users could splurge on afford them.
Data center QLC is a thing though:
https://www.anandtech.com/show/12423/micron-readies-3d-qlc-nand-based-datacenter-ssds-for-nearline-storage
It does have good hosting applications, i.e. for serving lots of static files. Much faster access than HDD's and cheaper than more durable SSD's, just not built for lots of write cycles.
my feeling is you should ditch one of your AX51 with ECC ram and move all data/clients to this one.
you get ECC ram directly and double cores and much more space here and you don't pay double price anyways.
the only issue with hetz is ip costs :'(
so if you divide it into 8 pieces... each can get 4/8 cores (depends on how you define a core) , 16GB ram and ~475GB storage which would cost around 12.5 and ip addresses extra
so with this server....you can offer better specs at your current price (or you can even offer small plans but fk ip costs)
ThemeForest Profile
Anybody wanna check my arithmetic?
What happens if we divide the AX101 into 32 slices, one for each thread?
Open issues:
Bad news first. Grumpy, ignorant, clueless administrator wants profit. ?
No RAID. No backup system. Could cut disk space in half and use RAID 1. Or zfs, etc. Or add more NVMe or rust at higher cost.
Almost enough IPv4s! Probably at least two people would want to stack other resources while needing only one IPv4.
Additional costs: Hetzner € 99 once off setup fee; Hetzner IPv6 once off fee; payment gateway fees.
Might it make a difference in performance if I switch from Ax51 with AMD Ryzen 7 3700X to AX101 with AMD Ryzen 9 5950X?
That's a nice price if user needs a lot of NVMe.
Fixed that for you:
Grumpy, ignorant, clueless Greedy administrator wants profit. ?
Greedy Hetzner € 99 once off setup fee; clueless Hetzner IPv6 once off fee.
RAID is for weaklings.
If the disk fails, just do 1 push-up for every GB lost to ask for forgiveness, and then move on.
RAID and backups at the same provider are rarely useful because the ignorant administrator could delete the data with a fat finger.
User should always keep a backup at a different provider.
We accept Karma donations for the last flan. 🍮 affbrr
Ordered one myself. Will do benchmarks and maybe some free demo VMs. However, i wont sell from this.
My little hobby site -> www.RaveX.eu
Install windows on pretty much any vps or dedi! -> Click here
Enjoy! ?
"The imitator dooms himself to hopeless mediocrity." — Ralph Waldo Emerson
the table above allocates one full thread per VM, presumably almost dedicated. if that's the plan and model, all good.
or, maybe it is possible to fit in 40VMs with say, 3GB ram. And maybe it's good to make RAID1 as a minimal form of protection? or adding 2 more NVMe to do a RAID10 to add value, speed and space. anyway it's still a healthy ~70-80GB per VM.
on top of that, might be worth considering allocating 2 vCPU per VM to allow bursting with an AUP (say 50% of a vCPU for sustained use etc)
with such a high single thread score, for reasonable installation / compiling use it would take much less time compared to say...lemme guess.....intel? so then the overall load in theory should be lower as everything gets done faster, so taking time into consideration its probably a lot less overlap/concurrency with neighbours' use down to the ms (or ns)? translates to more "carrying capacity" per node.
at some point contention ratio may have to go above 1:1.
I bench YABS 24/7/365 unless it's a leap year.
32 slices = you're supplying yet another small cheap KVM product. There's tons of that already. I'd be interested in something 1/4 of such a machine with a way of cpu-bursting to the whole machine, especially with added HDD. I don't need any ipv4 or NAT (pure v6 is fine since it's for private infrastructure).
How does this sound: 4 slices, 1 dedicated core (2 threads) per slice, 12 shared cores with soft cap of 20% max utilization of the shared cores per slice on a monthly basis. That is, you have 1 core to yourself and can burst the other 12 up to 6 days a month. I think I could make good use of that. I currently use a 4-core i7-3770 and run it at 100% cpu for a few days a month and it's idle the rest of the time.
Hi @willie! This sounds like a most interesting idea! Would you please say how much you might want to pay per month?
Without taking anything away from @willie's excellent idea above, please suppose that
There is an AX101.
it runs Debian on the metal.
Neighbors get an old-fashioned user account on the metal, but with enough sudo privileges to set up their own LXC containers or KVM instances.
There are only a few people on the machine. All of them are awesome.
All resources are "fair use," which means that some of the people can pig the resources some of the time.
Price? How much might you be willing to pay? Please consider prices of zero (free), $10, $20, $X>20 per month.
Anything else?
Who would come aboard this AX101?
Thanks very much! ?
I like your plan and @willie idea is also very good.
probably fit 2 big players and several small players (who mostly idle it) and price wise, current price coming to $4.73 so if you double it....you can still offer it at ~$9-$9.5 and ppl still get excellent performance and specs.
and i can jump in, if you
I am, with above small plan (or 2x of it)
ThemeForest Profile
Just provide the two disks as "passthrough" on partitions main disks so people can roll their own raid0/1? (not sure if possible and feasible)
Also, if you can get this stable, I'd take a 8-12gb thing for 15 euros/month. Would be cheaper per GB than their cloud line.
Hey teamacc. You're a dick. (c) Jon Biloh, 2020.
I like the old school time-sharing idea although I have more than enough resources for now.
I should not have used the term "fair use." "Fair use" often means some complicated percentage of the time policy that ends up as "not really dedicated."
What I meant was that each neighbor's stack of one or more slices can have its group of 100% dedicated "hyperthread nproc cores" ("ncores" = output of
nproc(1)) pinned, perhaps withtaskset(1), if that's what the neighbor wants, but usually provisioned via by the Linux kernel's very efficient default mechanism.No way do I mean that a neighbor who buys dedicated gets less.
¡Saludos! ?️
Thanks guys!
We will miss you! Maybe you can come aboard when the time is right. ?
Oh Tom... Oh @Not_Oles ... :
Not one but two... uber drooling.
Ooops! Yeah, I don't know how that happened. I meant to order only one! I wrote to server-order asking them for only one. Sorry! My bad!
Would around 35 euro for 1/4 of the base machine plus whatever for HDD space (would need other users wanting that too I guess) be reasonable? Are you still thinking of doing a separate storage server? I'm a little bit scared to commit to this just now, even though it makes sense for me in principle. I'm currently paying 28 euro for an i7-3770 with 32gb ram and 2x 3tb hdd's. I don't really care about SSD for what I'm doing, but I'd like ideally around 5TB of raw storage, preferably raid.
Does anyone know yet the make/model of ssd's on the AX101? They should be visible through hdparm -a or whatever.
From this comment on the OGF it is TOSHIBA KXD51RUE3T84 and it seems pretty solid (TLC per the specs and 1 DWPD - so enough and more endurance that you shouldn't worry about it!).
Why would you share resources for 35 euro. You can get ax41 for 34 euro (or 39 in germany) and move on.
How bout 4vcores (shared), 4gb ram, 50gb nvme with raid1 for $6? Thinking of doing that to my own server. Also, ill make it stackable. @Not_Oles should do something similar
My little hobby site -> www.RaveX.eu
Install windows on pretty much any vps or dedi! -> Click here
Hetzner fixed my mistake. Thanks!
So, one order remains pending. . . .
It certainly would not be unreasonable.
I might even take that offer if you still want to do it when we get to the right time.
That new SX server seems really nice. I would love to have one or three. . . .
But not yet ready to move in the storage direction. 
Interesting! I think your i7 is a great deal!
Your feeling a little concerned about commitment tells me that you are a reliable person, that your commitments are real, and that your commitments -- even informal commitments -- should be respected.
But, friend, no worries! We will see how everything works out, and I am sure everything will be fine!
@RaveX Thanks! What you are suggesting kinda looks like what I have been doing. So, sure, I could do it some more.
Nevertheless, something new might be in the air?
I kinda like @willie's suggestion of selling neighbors a quarter of the box. Ya know, it's peaceful to live in a small town with great neighbors.
Then there's @yoursunny's suggestion about hourly billing, automagic with an API. I wish I had @yoursunny's list of the minimum API features needed. Perhaps:
The remaining one order is a mistake too, consider that half of your last server is still idling.
Hardware has to be 80% utilized before ordering more.
We accept Karma donations for the last flan. 🍮 affbrr
The attraction is way more cpu speed (13 cores I guess), some of the time. Also it would include 2TB of SSD instead of 1TB. And I'd want DE location. I think you are right though, ax41 may suit me better.
My i7-3770 was a great deal when I got it 3+ years ago but current stuff in the auction is even better (my i7 is also an auction server).
Off topic, but have you had problems with the i7 crashing? I'm on my third one now and they all freeze up (almost) every day.
I had a E3-1245v2 from their auction that was excellent and sort of regret letting that go in favor of the i7, but needed the extra space (4x6TB).
FreeVPS.org - Grab a free VPS! | SOGo Webmail - Powered by Alinto
Haven't had crashing per se. Had some weird compute errors that looked to me like bad memory, so I opened a ticket and their techs swapped out the dimms. They say they do that on suspicion of errors. I guess they put them on a memory tester after swapping, and cycle them back into their servers if the tests pass. Also had a bad hard drive and eventually convinced them to replace it, but only after its error counts were visibly increasing. They were more reluctant to swap the hdd than the ram.
Wow! I'm all excited! Hetzner delivered the new server this morning, and I could log in!
I ordered some extra IP addresses.
The next step is to install. I will try a few different setups over the next few days. I'm thinking about trying zfs, which I never used before. The Proxmox guys seem to like zfs.
A couple of guys have mentioned possibly not using disk redundancy in order to increase the amount of disk which could be assigned to each slice.
Over the years I've experienced three sudden disk failures, including my first, where the disk suffered its fatal heart attack right in the middle of making a backup!
In the second incident, both RAID disks in the server died at the same moment, allegedly due to a power surge!
The third time was the easiest. Smart tools detected an irregularity, the disk was replaced, and everything was fine.
But now I have NVMe instead of spinning rust. Here's a quote from 1MachineElf on HN:
Of course, somebody else posts right away that 1MachineElf s wrong.
I'd feel better if folks who want the extra disk space would tell me how they do their own backups so that I can imagine, in the event of a server disk failure, they won't care. I had multiple backups when that double disk power surge failure happened. The provider offered everybody six months free service. I declined. I had my backups so I was okay, and it seemed that what happened to the poor provider could happen to anybody.
A couple more things:
If somebody wants an account to help with testing, I might be able to do it.
My AX51 in Helsinki might be up for transfer after the existing services are migrated. If you might want it, please feel free to let me know.
Now it starts to make sense. 66% more costs, 100% more resources.
In 2019, after spending two days installing 5 new servers, I powered off 3 and kept 2 online. A week later, one server had an NVMe failure. The NVMe device just disappeared from the system, invisible from BIOS or Live CD. I guess there's electric fault.
We haven't paid the vendor at the time so they replaced the defective drive swiftly, but I had to redo the setup. I also tested the other 3 servers for two weeks, but they didn't have the same problem.
List servers in account:
Create a server:
Destroy a server:
Uptime is in seconds. Price is dollars per second.
We accept Karma donations for the last flan. 🍮 affbrr
Yes, RAM is up 100% (64 to 128 GB). Notably, disk is up 340% (1 to 3.4 TB).
Also 30% faster single thread cpu benchmark.
This is a great story! Thank you for sharing!
Got it! Thanks!
@yoursunny We joke around so much, and there is great fun! Perhaps also, sometimes, some small risk of misunderstanding might arise.
Although I appreciate that you have lots of resources for your research and for your website and other projects, I nevertheless jumped to the conclusion that you actually might want to use my server for something sometime.
When I was joking about the API increasing the push-up count I was ineptly trying to suggest that payment might be adequate in push-ups and jokes instead of dollars. Maybe one push-up per processor day, or something like that.
To me, your research seems worth being supported even by little folks like me who are incapable of endowing a university building and a professorial chair.
I thought I was humorously masking my possibly good intentions in what perhaps could have looked like sarcasm.
If that happened, I am sorry!
Now I will just leave this topic by suggesting that, if ever there is some little thing that I could do for you on one of my servers, it's okay, but you probably will need to help me implement whatever it is that you want.
Best wishes from Mexico! ??????️
I don't see any misunderstanding here, and I do not interpret sarcasm. Thus, what you might fear has not happened.
It's possible to design an API that supports multiple currencies, but that would not be minimal.
In SolusIO, the native price unit is token, and the provider can set the exchange rate between token and dollars / push-ups.
I don't have a chair. I place the computers on top of cardboard boxes on top of the table, and I operate the computers while standing. It's healthier this way.
We accept Karma donations for the last flan. 🍮 affbrr
One suggestion, in case it hasn't crossed your mind (yet), is to have both RAID 1 (mdadm) for your main OS (avoid ZFS only because rescue with ZFS isn't very straightforward). A simple <50GB partition should do the trick for Proxmox and friends.
Have a separate RAID 1 setup (either mdadm or ZFS) for the OS for your VMs - this can be a smallish size that is given to your "customers" (think UltraVPS, who typically give you a 10GB partition to install the core OS - maybe you can bump up to 20GB depending on the size/purpose of the other allocation for the VMs). You can then have a separate RAID-0 stripe of the balance disk which is the "non-RAID" storage given to the VMs at their own risk.
Best case, 1 drive dies, RAID-0 goes and everyone knows but they can still boot and rebuild data.
Worst case, both drives die - nothing you can do except restore from backups.
My thinking - 50GB for your Promox host (RAID-1), 200GB-600TB for RAID-1 for VMs (depending on demand the size will change), Rest of disk ( > 2 x 3TB ) - RAID-0 stripe as additional storage for the VMs.
Profit
Best wishes...
This reminded me of https://web.archive.org/web/20060423082708/http://www.fedexfurniture.com/
I thought about adding more disks. Interestingly, additional 3.84 TB NVMe disks seem to cost € 48.00 monthly each. Adding two would cost € 96. An entire additional server seems only € 3 more than simply adding two extra NVMe disks!
I thought about adding one or two spinning rust drives. But the added cost seemed more than I wanted to pay.
Right now the server is up with ext4 and RAID 1 on its two 3.84 TB NVME disks. Making 30 of the 32 logical cores into slices and giving each slice 100 GB might work. zfs and friends will continue improving between now and next time.
Hetzner has some thorough tutorials online about installing and configuring Proxmox. See here and here. But Proxmox also might be more easily available via the Rescue System's installimage installer, the code for which seems to be on Github.
We will see.
Much appreciated! Thank you! ?
I wonder if OpenBSD will work. One of my friends wanted to try doing the install, so I got it ready for him.
Lots of fun!
What was I thinking... maybe someday though that looks too far fetched for mere mortals like me. Though @lentro will disagree...
That makes sense. If you literally divide the server into 30 "chunks" evenly, netting 100GB NVMe + 1 vCPU + 4GB RAM at roughly $5/month (including 1 IPv4) will be > break even for you cost wise and it is (I think) a good VM to have considering the resources (assuming fair share and that there are no abusers). I think though, that this is still a "typical" VM with nothing outrageous resource wise. As others have mentioned, I think you'll find more takers with some unusual resources like 10+GB RAM/4 cores/disk etc. appropriately priced because this kind of a big honking server is more suitable for some serious loading.
Either way you have exciting times ahead to test out various configurations and options. It's a wonderful problem to have.
Good luck and it is invigorating to read on your experiences as you discover more...
Thank you for sharing.
Yaaaassss. I am now implied as immortal!
So far all the neighbors on my servers have been really nice! So lucky!
Hopefully it's a little faster than some others?
1 vCPU = neighbor's choice of either 1 100% dedicated logical core or 4 x 25% burstable to 100% logical cores.
I need to do some arithmetic on the pricing and also post a yabs benchmark that I ran yesterday. It's getting a bit late here. ? I am looking forward to another happy day tomorrow!
I wish the same to all of you! ?
I know it can be installed on AX41/51/Intel E3s on proxmox so no reason why it shouldn't! Installer isn't very flashy and the partitioner takes some getting used to lol