Contributing with hosting

[kristosh]

Hi!

Not sure where I'd post this, but figured FAH Hardware wasn't too out of order.
Please move the post if you feel this is more appropriate in another category.

I do some work with war on Team Hardware (37651), and know that with the recent COVID-19 news, there is a lot of traffic on the assignment- and work-servers set up for F@H.
This is great as it definitely means a lot of people contribute to crunching through the jobs, but it means that we often have hardware standing still as they cant receive any jobs.

I work at a server/hosting-company and have access to lots of hardware and connectivity, really just looking to see if I can contribute with any hosting-tasks to help relieve some stress from the official hosts.
I haven't looked to much into how the tasks are generated and exactly what the work-servers do (except from distributing), but I feel I understand the assignment/work-server idea.

If so, what are the minimal requirements to host such a service?
We can easily scale to many different needs, and if I raise enough interest for this within the firm, we could probably dedicate a host for this permanently.

Thanks!

[Joe_H]

I will pass on a link to your post, someone should get back to you.

Not sure of the requirements, I do know at least one of the Assignment Servers is hosted on a cloud service. Work Servers are a bit more of an issue, terabytes of storage are needed. Older ones have 10's of TB, the two newest ones had over 100 TB, but that is filling up.

[kristosh]

No messages yet, still eager to help with the increased traffic if it could help at all.
We are having serious issues receiving any work at all for our multiple clients, so we are very interested in helping our the community.

I'm not sure what would benefit the most, but for a limited amount of time we could probably put up of our larger spare-hosts (sub 100TB) up for temporary hosting.
For a permanent setup we could probably go for either a speed-oriented setup with 15K drives (4-5TB after redundancy) or possibly a 7.2K setup with more capacity (++20TB or so after redundancy)
This is really just a side-hustle for our part, so please understand that we can't dedicate our newest hardware for the task.

Any insight into what is the bottleneck with the official servers? Is it the assignment servers that are overloaded, or is it the work-severs that receive too many requests?
We have multiple 1 and 10Gbit connections to various Internet-providers, so traffic should be OK infrastructure-wise.

[Joe_H]

Any insight into what is the bottleneck with the official servers? Is it the assignment servers that are overloaded, or is it the work-severs that receive too many requests?

As I understand it, and am just a volunteer here, the AS's can handle making more assignments. The WS's have more limitations as each connection involves staying alive longer to do either a download of a WU or handle an upload. The uploads are direct from the clients, so are not mediated by the AS's. More WS's are in progress, two based on Azure just went into service yesterday as shown - https://apps.foldingathome.org/serverstats

[toTOW]

Any insight into what is the bottleneck with the official servers? Is it the assignment servers that are overloaded, or is it the work-severs that receive too many requests?

The main bottleneck is the disk IOs on the WS. Most of the WS are designed to store a lot of data and under a standard load, it doesn't require a lot of disk IO.

I'll try to wake up someone to answer your offer.

[sukritsingh]

Hi! Thanks for your patience and support! Please shoot me an email at sukrit.singh@wustl.edu so we can connect and discuss what your hosting/server capabilities might be! At the moment any and all help on that end would be invaluable!

[kristosh]

As I understand it, and am just a volunteer here, the AS's can handle making more assignments. The WS's have more limitations as each connection involves staying alive longer to do either a download of a WU or handle an upload. The uploads are direct from the clients, so are not mediated by the AS's.

Yep, that's what I thought. Good thing the AS's are keeping up with the traffic then.

The main bottleneck is the disk IOs on the WS. Most of the WS are designed to store a lot of data and under a standard load, it doesn't require a lot of disk IO.

I see. In that case a smaller (<10TB), but faster host would help offload some traffic, right?
How often do jobs get uploaded to the WS? And for how long is current work stored on the host?

Hi! Thanks for your patience and support! Please shoot me an email at sukrit.singh@wustl.edu so we can connect and discuss what your hosting/server capabilities might be! At the moment any and all help on that end would be invaluable!

Just send you an e-mail, we could get things up and running rather quickly if we could settle down on the configuration.