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.
FOSSVPS.org is born!
This discussion has been closed.
Comments
You shouldn't assign whole 128GB just for KVM's. Always leave few GB for OS and cache.
8 IPv4 plus IPv6 VPSes, each with 4 vCores and 8 GB of RAM.
another 8 VPses with IPv6 only. Also 8 GB RAM. + NAT4 outbound interface, it's a one-time setting.
this is a perfect plan.
Sounds good! Thank you!
I imagined that people might take part of that RAM and make a very fast RAM disk. Just one idea.
Thanks for your comment!
Excellent advice! Thank you!
I think you would want inbound NAT IPv4 also included?
Thanks! Sounds good to me!
Thanks to everyone who has commented above!
Maybe, for the first round of 8 VPSes:
CPU: 2 vCores
RAM: 8 GB
Disk: 69 GB (1100/16=68.79)
IPv4: 1 x /32
IPv6: 1 x /68
Bandwidth: 10 Gbps shared
Location: London, UK
OS: Ubuntu 24.04 LTS
Possibly a second round of 8 IPv6 only or NAT IPv4 plus IPv6 VPSes.
What do you guys think about this configuration?
Thanks to @dan_onlyservers and @netixen for the nice server node!
Looks solid, from when it's starting?
I for one, prefer when my hosts overcommit memory
The 8 initial VPSes need to be created with macvtap network connections. I don't have much experience with macvtap. So stating will have to wait until the VPSes are created and tested. Hopefully not too long!
Mostly I have made VPSes by directly calling qemu from the command line. Recently I have really enjoyed using BashVM and @babywhale has been super helpful. BashVM makes "automatic" and "manual" VPSes. "Automatic" creates NAT VPSes, while "manual" creates, I guess, and kind of network.
A fun fact is that BashVM's NAT VPSes work with the qemu console while BashVM's manual VPSes use VNC. I want the Qemu console and not VNC. So, right now, I don't know how to make a macvtap VPS with BashVM and have that macvtap VPS use the Qemu console. I've been discussing this with @babywhale so we will see what happens.
Meanwhile, I've been having especially a lot of fun today, the first time for me making test VPSes with cloud-init.
Coming up, @babywhale might help make the VPSes. Or, maybe someone else also with amazing skills will help out. Or, I will keep working on the VPSes until I think they might be okay.
You will eventually run into IO issues if doing 16x69GB even if thin provisioned. Doing so and hoping users do not use their full storage allocation is like driving a car with a blown head gasket.
Allocating all available storage to VMs will leave nothing for system overheads, backups, ISO's and images, logs, etc.
A more sustainable approach with your available resources would be thick provisioning ~50GB per VM (if deploying 16 instances) and if you wanted to ensure continued performance, then consider 6/7GB RAM instead.
Thanks for your comment!
Overcommitting memory actually seems to work, at least initially, when the memory that's used, as oppsed to the memory that's allocated, is within the available hardware amount plus swap.
Adding some swap on the node and on the VPSes seems reasonable, except for the issue that the swap file can keep sensitive information cached.
I haven't tried over committing memory enough to arrive at a fault, for example an OOM fault.
I haven't studied the code enough to know how the memory allocations are handled and how the actual memory in use is determined relative to the allocations and what is being run. It seems clear, though, that mere overallocation is not itself enough to cause an immediate problem.
@beanman109 Do you know the qemu and the kernel code well enough to provide insight into how the memory allocation works? I would be interested in anything you can teach me. Thanks again!
Sounds correct! Thank you!
Maybe we could resolve this issue by deploying less than 16 instances. For right now, the current plan is to deploy the first round of only 8 instances.
RAM: 8 GB
RAM: 6 GB
Disk: 69 GB
Disk: 50 GB
Sorry all I know is when I sell VPS if I have 128GB of RAM I can fit 512 VPS with 1GB on that, complaints about VPS crashing normally only start coming in after the 2nd month and at that point I've long made off with the money
Congrats!
FWIW I didn't add the swap. It was already present. I might increase it at some point.
Thanks again!
I think 2 cores might be a little too less. This could be compensated by aggressive cleanup of VMs that are high consuming but 4 cores would be nicer
@Protocol903 I get your point that 4 cores are nicer than 2. The view I sense in the wind is that people don't want overselling. For example, what would you think if, when you started using your four cores a little more, your account was aggressively cleaned up? Maybe I'm wrong, though. Best wishes! Tom
A possible solution might be to make 16 automatic NAT VPSes with BashVM, and then manually allocate the first 8 to macvtap and their own individual IPv4s.
Hmm. Thats a fair point. It would be very difficult to be fair and find a sweet spot in antiabuse cleanups. How does 3 sound? 8G3C would fit perfectly fine for a medium-low application that dont require heavy CPU operation. Hell, 6G3C would be nice too but if I were you, I would personally stick with 8G3C to discourage people from using swap to compensate for lack of RAM.
Low IO speed is more painful than slightly high CPU steal.
Nice!
Excellent! Now there are 16 VPSes ready to go. The above yabs is from inside the first VPS.
These VPSes are IPv4 NAT plus IPv6. They each should have 20 IPv4 ports, one of which is assigned to ssh.
Maybe the first 8 can be adjusted from NAT to macvtap. . . .
When these VPSes are distributed I need to remember to turn them on and to enable autostart.
Getting closer to launch. . . .
Which OS/distro will they be running?
Nice YABS you've got there 🔥
@BasToTheMax
Thanks to @dan_onlyservers and @netixen for the kind server donation!
Thanks to @babywhale for BashVM!
If I (ever) get one, would it be possible to install alpine (or debian)? 😉
I tried installing Debian 12 with BashVM. On this node, running Ubuntu 24.04.2, here seemed to be network issues with Debian 12 VPSes. It's probably some simple mistake that I made.
Also, I don't remember seeing Alpine listed among the OS choices offered by BashVM.
So the answer is, "Not right now, but maybe a bit later."
Reminder to myself: Make a set of backups of the original installs.