Co-operatively owned server hardware

Slighted off key and on key. There is a fundamental mistake in the operation and attraction of capital for this kind of service. Coops can not be built in the current way they are funded. I will come up with solutions and ideas. Presently the global economic outlook for the future is not bright so investments are scare from the public individuals as we head towards a great depression. I see data centres value be diminished very quickly. Since large operators like Cisco try to compete. There will be alot of competition eroding alot of smaller data centres or making them be purchased from largers giants. Though coops have a huge advstange in this arena I simply need to develop my cooperative model and its article to file at a regulator that wont swipe us away from the fold of commercialisation and is good to attract capital from so anyone no of any jurisdictions that have good laws but fewer regulations some but flexi. I welcome feedback

We recently brought a new file server, it isn’t racked yet, due to the lack of capital we opted for the cheapest, minimal options for a machine that could replace our existing server and also give us a serious increase in disk io, reduce the electricity demand and also some additional space, it is a 2U server with 20 2.5" disk bays, 8 populated with 1.9TB SSDs (so there is room for more when we have the funds) and this cost about £7.5k.

3 Likes

Great - I mean not so great the cost, but at least we have a number!

The application servers, what sort of price tag do they come in at? The sort of front facing things like ths very forum is hosted on?

2 Likes

The next front facing server we are looking at getting is a AMD 2nd generation EPYC 24 core CPU (48 with hyper threading) and 512GB of RAM and from memory this will cost in the region of £4-5k.

1 Like

Wow, thanks for sharing @chrisc.

:exploding_head:

Sorry for the very basic questions.

I know it depends on their needs, but roughly how many sites would you expect to host on that machine?

Obviously the second rung is complete cost of ownership here, with includes paying for someone to maintain it.

Not asking in the abstract here. I’d genuinely like to consider whether a few coops should do this, pertaining to this other thread on a plausible Heroku alternative.

I’m guessing that you are asking about the front facing server?

For the VMs that we provide services on we resource them based on need, for example to avoid swapping we allocate a lot of RAM we also don’t want them to fall over with the occasional load spikes, we give them CPU cores based on need, the big ones:

  • Mailcow, webarch.email our main mailserver, has 36G of RAM (it had 48G last year and looking at the stats it probably needs to go up to that again as it is now swapping) for 275+ domains and 1,200+ mailboxes
  • GitLab, git.coop, has 20G of RAM and 830+ projects and 150+ users in 85+ groups, the corresponding GitLab runner Docker server has 16G of RAM and 8 cores to support 8 concurrent pipelines.
  • Shared web hosting, we have 10 shared webservers, these need to have 16G of RAM each to ensure that all the PHP can be kept compiled in memory with OPCache and the MariaDB InnoDB tables are also held in RAM and also enough resources so that if one or more sites suddenly has a load of traffic the server stays up. These servers have up to 50 users accounts each with a mix of static HTML site and PHP / MariaDB sites.

Client VMs very between 1G and 10G of RAM each (for example community.coops.tech has 4G) and as clients pay directly for them they are often somewhat under resourced to save costs, because, like others, we charge based on RAM and disk (we do allocate CPU cores based on need) and although you can’t get away with too little disk space you can often get away with too little RAM and thrash the disks via swap to make up for the lack of RAM — this must be happening all the time with providers of cheap VMs.

In terms of disk space, once you have datacentre grade disks, RAID, snapshots, enough spare capacity that the array runs well (once you get to 80% capacity things can start to go wrong), a backup server with 30 days worth of backups and a backup of the backup server in another data centre with 30 days worth of backups then the cost per TB of storage no longer has much correspondance to the cost per 1TB of consumer grade disks…

The reason we allocate CPU cores based on need is to try to get the most usage out of the hardware without under resourcing or over resourcing virtual servers. We never go over two virtual CPU per hyperthreaded core and aim for one-to-one, so a 512G RAM server with a 24 core CPU (48 with hyperthreading) equates to virtual servers with around 10G of RAM per virtual CPU.

We have found that although servers generally only come with a three year warranty that after three years any hardware problems have been solved and they can be used for a further seven or so years, in a variety of different roles.

It is also with noting that the cost of of hardware, rack space, electricity and connectivity is very much related to the quantity purchased.

It is a capital intensive business in which you are trying to compete with capitalists, when you don’t have any capital, within a capitalist system, with clients who are often asking why the prices and the services of the capitalists can’t be matched… as I said at the top of this thread almost two years ago:

And in the third quarter of 2019 “Amazon Web Services… ended the quarter with $9 billion in revenue”, and for the first quarter of 2020 “Amazon Web Services revenue was expected to come in at $10.33 billion”, so for 2020 AWS revenue is probably going to be over $40,000,000,000 USD.

We have intentionally decided to host the first meet.coop server with Koumbit because they own their own hardware… a couple of the key points of co-operatives are “4. Autonomy and Independence” and “6. Cooperation among Cooperatives” … I find it hard to understand how building a co-op on the back of Amazon Web Services or other commercial cloud provider fits in with this…

3 Likes

For me the rubrick is not “can we genuinely compete with Amazon” (or Google Cloud or so on) which, we can’t. But can we builld a compelling enough service that encourages cooperatives in the medium term and others in the longer term to make the shift. Which you know because you and I have been talking about this for four years!

Most coops are not using AWS or others because they really want to. They are doing it because they think they have to or do have to, for a variety of financial, operational and organisational reasons. I’d be interested to survey CoTech members as to why, if this has some legs.

For me personally though my task is, I’d like to replace our use of Netlify (useful but expensive for glorified web serving plus continious delivery) and Heroku in the next year, but simply haven’t had the headspace to consider this until now. We don’t use AWS itself, but these do and we’d like to move on, especially as Netlify is ridiculous when you look at it coldly.

2 Likes

We run our own kit.
We also use Hetzner or similar occasionally, but the vast majority of our stuff is on our own servers.
I’d ask the sysadmins to provide some more detail but they are busy working on the database server right now :rofl:

1 Like

Kit and colocation - Systems - The meet.coop Forum :bomb:

2 Likes