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
we might actually have first madman config ready this week.
Hardware from Entropy.
The Process: Slap it together, break it, refine.
Purged In Flame, Reborn In Uptime
(Yeah yeah... typo in the title. But the fire’s real.)
Chaos reigns as progress happens, servers emerge from the entropy.
Another day at the lab (or madlab? chaos engine?). Posted some in our discord as well for your guys to see.
Before uptime, there is fire.
ePA-CF power control module cleanup. No matter how long you dry at 70C you still get these -.- Also CF, not good for your health. Got 20kg in the shelf here, but not certain we will use it. Carbon Fiber dust is a bit like asbestos ...
3D Printed M2 Threads — And Yes, They Work
Just need a printer with tight tolerances and right slicing settings so that slicer even tries to do this small details.
Not only that, but these were printed VERTICAL, not horizontal. That's the hardest way for printer threads to function, and by all sense shouldn't. But they do.
Tested on ABS, ePA-CF for this model, done on PLA on others. We regularly do 3D Printed threads, just make sure you have 3+x of diameter for thread engagement.
Sacrife To The Server Gods
🪦 RIP N100 NAS Motherboard
🧠 Uptime: 0.0 seconds
💀 Cause of death: 24VDC straight into PicoPSU
👨🔧 Diagnosis: Human
🏆 Medal: Lab-grade Darwin Award
One moment it was full of potential.
Next moment — full of volts. Booted straight into Nirvana.
Note: When connecting the lab bench power supply to something ... check the damn voltage first. Turns out, these little guys don’t like 24V raw.
On the other hand, they run 48GB of RAM
Booted straight to BIOS. Looks stable. OS test coming soon.
🧪 Building servers with entropy.
⚙️ Engineering from the edge.
🔥 Creating from chaos.
More coming. Join the Madman from Finland as we torch, build, curse, test, and post.
. The chaos is real. Join 
to see more raw photo streams.
Forgot to mention yesterday, this is one of the reasons why we stated 2 months; See the photo with SATA cables?
Well we were supposed to have the right lengths in stock, had already bought them etc. but turned out they were like ~5-10cm too short and the wrapping couldn't be opened either ... 1 friggin' tiny piece, and it all halts. Don't even remember how many times a 0.10€ part has stopped from delivering a node for weeks. You need every tiny piece. And that's how you get to the point you spend over 500€ just on HDD screws to never run out again! and at one point we used M3 instead of #6-32 knowing it will be pain to remove years down the road because we were out of #6-32 counter sunk ultra low profile HDD screws. Been fun removing hundreds of those.
Local supermicro dealer has some long cables in stock at exorbitant prices, but we need in volume so waiting for delivery from China takes easily 4 weeks.
... which reminds me, forgot to order the very short ones. Eh
This is HW Development at it's finest; 1 little piece missing causes 1 month of delay. No one ever sees that part, end users just click "buy" and it just works -- not knowing the pain and blood behind the other side of screen it took for you to be able to click that "order now" button.
Had 0.10€ part halt your project? Share it with us and get some service credit!
Oh yeah THIS very first prototype could be ready to go live tomorrow night during the electrical maintenance, will be there personally fiddling my thumbs for hours during the night to ensure everything goes smooth. Comment below if you want to be the one to test one of these units, we'll make a sweet deal for short period if you share temp data etc.
@PulsedMedia Will rutorrent provide performance with such a weak processor?
Uhm, rutorrent would run on a Atom 330.
rTorrent is the client beneath, and even RPi 3 or something like that is more than sufficient.
Infact, our original production servers were Atom 330, 525 and others low power of the day 15years ago. rTorrent has barely changed.
With N100 we are talking about i5-6500 performance at way less wattage required. That is in no way a “weak” processor. imo considering the target market, you can throw anything at it and it will run them decently.
I have a smol N100 mini pc too and it’s crazy the value you got for the money.
interesting I don't think you can run more than one torrent
You can download all the variants of linux in paralllel
And decoding av1 or hevc in the meantime with integrated gpu.
This cpu is great.
The Intel N100 is excellent!
6W, modern, built-in iGPU ...
We still use these old opterons in production 2419 EE, here's passmark comparison:
CPU Perf
Memory
Results
The opterons sit 95% idle in seedbox usage. They are 16 years old, and by CPU benchmark metrics completely freaking useless. N100 offers ~5x single thread which for rTorrent especially is more important (mostly single threaded), it's the latency how fast something gets processed to get stuck into the I/O queue.
It's all about the drives beneath to be honest, nothing else.
The opterons get bogged down on modern SSH ciphers tho, since they don't have HW acceleration for the new heavy algos. Easily solvable by using slightly older ciphers which are still strong, but alas, these opterons are on their last fucking leg. We squeezed every single dime of value out of these units.
We have customers running 15 000+ torrents on single rtorrent instance on these machines for more than a decade now. On a shared server, with other users, up to ~16 on single server, but typical more like 8. Some of these guys probably bought a 100Mbps service on a Atom originally back in 2010-2011, so they've got already massive upgrades over the years (+bonus storage quota over the years probs maxed out). These are all legacy plans.
N100 Utterly obliterates the 2419EE in every meaningful metric. It's not even a friggin' contest. It's a gladiator vs. skeleton buried half a decade ago type of contest.
These Intel N100 units would be a massive upgrade for those on opterons still, and actually enable HW transcoding for streaming, no more just direct stream but also transcoding for users on these units.
Mind you, we haven't setup a single "new" opteron 24199ee node in probably 4-5 years (new means refurbished and new drives in this instance). We've had most of these units for ~12 years now. You can still buy a service which runs on these nodes, and we've yet to have any CPU performance complaints on these, only about I/O bottlenecking to the drives. It's always the I/O which is the limiter. rTorrent is insanely efficient. ruTorrent is mostly hard on the client browser, not the server.
ffs, these units "died" once about 8 years ago, we breathed new life onto them with some elbow grease; DCS made a mistake, we fixed it with 40mm fans on the chipset. And then we just kept running these units.
In practice a 2RU 3 cloud node setup with 12x3.5" consumes about 180-190W, with full activity. So going with these N100 would double the density and increase node count, WHILE increasing the raid array size from 4x3.5" to 6x3.5" giving a storage efficiency boost from 75% to 83.33% at the same time too, and offering BW upgrade from 1Gbps to 2.5Gbps. That 2RU power consumption we are expecitng to increase to about 300W from the current 180W tho, but 12.5W per HDD, instead of 15W per HDD. But only true productio nwill show. Better yet, if we make these seedboxes, we get 48 users per 2RU instead of 24, and 6.25W/user instead of 7.5W.
NVMe could be used to cache, but the reliability metrics with bcache .... Yea those ain't good! Infact, bcache is very bad for our uptime metrics. With these it might work (small enough service domain)
rTorrent is very single thread limited software, it is very very efficient, but very single threadish. Deluge, qBittorrent etc. can leverage multiple cores, but they are heavy AF compared to rTorrent.
btw, we just restocked some of the MD units and they are now lower price for the moment than in a long time. Maybe grab one? People use these as seedboxes + then mount with rclone+caching one the storage boxes for bulk storage. Setup plex etc.
Yes, the rtorrent + rutorrent interface that needs to be explained is easy to use, the only problem is that it uses a single thread of the processor.
See my post above, exactly as i posted.
You might want to try reading the post in full.
The Intel N100 is not a bottleneck, if you manage to make it a bottleneck you are configuring something very very very wrong. If you can run rtorrent with 15 000 torrents on a Opteron 2419EE with fraction of the performance, the N100 definitively will not be the bottleneck.
my only bottleneck woth PM is the inability to reinstall your own servers. Its been 3 years now that 'its on the way'
Since 2015;
I have tried 3 pulsed media seed boxes before. they are all in a terrible state.
SSD Beta 100G ( cancel )
M10G S ( cancel )
Dragon-R Mushu ( cancel )
This is the situation, take care of yourself.
it does suck, i remember when i got a 2x 2TB NVME and support just refused to let me configure RAID0 which defeats the purpose of multiple disks
Yeah for as long as I've been on here the reinstalls are cominnnnng. The new interface is cominnnnng.
Not true, any distro is available for the high end servers. MD has not existed for 3 years, which is the lineup which is limited.
Albeit, we have never accepted Windows installs.
This is MD series, you can trivially configure RAID0 and ask through support RAID1/RAID0 support out of the box. The partitioning is made especially for this (separate /home). Self managed. I think the default is now RAID0 for bulk data, but we still enforce root to be RAID1 for what should be very obvious reasons (if another NVMe fails, atleast you know it failed immediately ...)
It's ~6 commands to change between RAID1/RAID0 (roughly; mdadm stop, create, scan&detail, update initramfs, update-grub, reboot). Most likely you were given instructions on how to do this. Self managed, but very trivial.
We did the Deploymentcode's TenantOS, it was not up to our needs at all, infact, it was in certain things much worse than NOC-PS. Albeit nice eyecandy, some very nice features, but on the backend? Absolutely horrendous to manage volume servers, and there are sections with zero documentation and you have to guess (we asked through tickets tho)
We've been using NOC-PS on the background for all mass produced servers, and for custom order servers it's manual process.
For the 10G Unmetered, storage servers, custom build, those which cost actual money, you just open ticket and request whatever you need (as long as it's Linux).
Oh almost forgot; You want custom OS? Just ask rescue mode and install how you wish. from a tar ball, rsync, debootstrap etc. whatever.
I still have to open a ticket for it. I wish I could just do it myself from a panel.
@PulsedMedia any seedbox deals? Looks like has been no deals since August 2024.
Sounds like you purchased wrong service if anything, or something else. Or maybe you just want to swap services periodically. Maybe you were looking for seedbox to race with. Or maybe you are just looking to flame. Looking at your account, you have only a couple of comments here on LET...
Whatever it is, the thousands of users who are happily using all of these services cannot be incorrect. Some of whom has been with us for 15 years, some of whom just a few months.
Our track record for providing seedboxes show that these services work very well for our customer base.
Not every service works for everyone. Maybe you need a Ryzen 9950X3D / Intel i9-14900k for your rTorrent instance, but almost no one does. These are not about CPU performance, these are about I/O. And sadly, as with everything Internet -> Some routes just do not work. Rural India, just cannot get high throughput, not even full 10G per thread to a 56k dialup! ... Joking ofc, but guess what ... We've even seen people who demand that. Infact surprisingly common that people expect single thread full link speed on entry level shared service to any target location.
Sorry we do not build seedboxes for racing, we build for the masses with steady regular use and high value, not for the glory of being the fastest on swarm for the first 30seconds, but for those who look for steady standard use.
Oh btw, SSD Beta 100G indeed was old, not even on our own hardware or network, maybe OVH i think? It was years down the line when we moved SSD services to our own hardware and network, it's completely different now.
M10G S is RAID5 HDDs, and Dragon-R is RAID10 HDDs. If you were looking to race, these are wrong services for that. M10G SSD maybe, but the tuning is not for racing either.
There are many use cases, racing has never been our core audience. Those who want to race should get dedis and custom tune their servers for their specific use case.
Performance issue tickets are rare for us, when they happen it's usually an abuser on a specific server OR 3rd party network issue beyond our control being the most common (say route to Asia, some ISP in between is congested upstream)
Reinstalls are typically handled within ~24hrs.
Custom distro install? Request rescue and for example take a look at our cloning script: https://github.com/MagnaCapax/mcxSauces/blob/main/baremetal/mdSeries/cloneLiveSystem.sh
It's for specific partitioning and debian(/ubuntu) but you get the gist.
Also people tend to run proxmox when they need something out of the scope. We got people using proxmox + moonlight for gaming from the MD nodes
my point still stands
I like you Aleksi but this is a major barrier. I dont want to open a ticket when I fuck up waiting for a reinstall.
Yes, and it takes something most precious; Human time.
Unfortunately, we can slap something shit together and spend shitload of time getting the templates to work -- OR we can do proper hardware and take the hard route and make proper system, MVP done in the same time as just getting 3-4 different distros working on something like TenantOS / NOC-PS, but new distros take like 60minutes to setup ...
Yea this is not good way to get the SATA power for our needs. Got to find a better way probably.
and few milk v duo s arrived ...
Forgot to mention, 3rd party proprietary solutions cost money directly AND in with slower backend stuff, slower to make OS templates (or re-do what we are planning regardless), limit what we can do (or take even more effort to make happen than doing fully in house) etc etc.
It's a trade off, pain now, streamlined future.
If you know of a system which automates a shit ton of the backend admin stuff, let us know. None of them do afaik, for most companies spending 30-90minutes just on the inventory side of a new server is no issue, when individual servers cost minimum 1000€ a each ... For us? Cheapest might be 100€ all-in!
We are targeting scale eventually, it's slow to get shit started and rolling tho
Other option is then that we will simply not sell these beta grade units at all. It's either sell and do limited SW and automation, or do not sell at all and this never leaves my garage (ehrm, lab) and remains a pipe dream.
So think of it as early adopter pain, everyone who can live with Deb 11/Deb 12 right now is part of making this reality at scale in so many ways. All of you are now part of the story
You have a long answer and an excuse ready for everything.
Majority of the providers here are able to provide that ability. My way or the highway isn't the only way to do business.
HW Development ain't easy!
First prototype almost done, already doing cable management ... when it finally catches my attention, why do the picoPSU feel so damn high?
Yeap, this is a showstopper. This is why 2months is needed.
Those are from top of ATX power connector to top ~21mm. There are no PSUs for ~10mm height.
So need an atx power extension.
These Mobos are no good for 1RU use then i guess. Need to shave at the very least 8mm.
printed adapters are 10mm tall ... 3mm height might work ....
No excuses, just the reality of shit from a small operator who is vertically integrated. This is just how shit works when you are building shit others don't.
Sure we can put 50k € down and hope a random dev house gets the job done half baked. But that 50k € is away from hardware then.
I understand the frustration, i am frustrated too as it would save time, increase sales, improve quality etc. So i do understand your point of view too. When something frustrates you, it feels 10x worse tho. Just ask for a reinstall / do whatever distro you want from rescue mode, besides how often do you need to reinstall an dedicated server anyways?
Or just get tenantos and bleed through every orifice down the road when it's not flexible for how we want to do stuff and bleed insane number of hours to get all the various templating stuff to work (which is poorly documented) or wrestle tenantos to flex into how we want to do stuff long term.
TenantOS is nicely built, but it's built for lower volume, higher cost services than ours. For us right now, NOC-PS works actually better than TenantOS. No joking.
To be real, how many do you see out there offering this power class nodes for this price, and vertically integrated DC, Network and even the friggin' hardware is built in house?
That being said, dev cost is "one time" cost, so it's definitively a priority, but first you need the hardware to run it on, and you need the numbers to justify the work.
It's not like we are just sitting on our asses here, but i personally got a choice; Either make the hardware, or make the software.
Hardware makes the moolah, software requires the hardware.
Therefore hardware first. Or nothing will ever get shipped. MVP, then polish, fix, enhance, repeat.
Any shortcut taking now is a nightmare down the road, and we plan to be running in 30years time, not be yet another summer host ... tho 15years of ops already behind. If we throw some jank in there, it's going to bite us in the ass later on. TenantOS is awfully slow on the sysadmin side for example. and we still would need to code most of what we are planning ourselves because tenantos doesn't have it.
Most others push out just prebuilt servers, they don't build their own hardware from scratch and have to manage all of this. They got IPMI/BMC for example. They got the price tag too.
But, if you got ideas, just let us know, we'll check it out.
So please if you got any ideas, do let us know. What you need for MVP etc. would be good, if you don't have ready solutions in mind.
We've hired 3 different coders in the past 1 year, from novice junior, to supposedly experienced hacker at higher education level ("just formalizing into paper degree"), and legendary level hacker (linux kernel has some of his code). Only the last one produced something tangible and useable, but he has too limited hours and other interests. Middle one was essentially fake credentials beta bob who couldn't hack it. The junior had typical school experience only; No framework? No hacky!
He would've been probably good as grunt level coder on web applications, but he also got another job which was closer to what he knew and interests.
We had one excellent coder few years back, Russian when the Ukraine war started so nobody wanted to hire him. He almost immediately found a corpo job with 50% higher pay so ....
That means right now, almost all the backend coding rests on my shoulders alone. So does the hardware design and production until everything is formalized (we are outsourcing some subcomponents right now tho!). Soi if i do the software, hardware doesn't get done.
If there is no hardware, you don't get to complain for lack of software features.
Oh and don't forget, electrical maintenance shortly. Everything might boot twice this night
Long time without seedboxs deals
@johndeo983 @rm_ no deals or specials on seedboxes right now, sorry.