[Jenkins-infra] A random thought for reducing ci.j.io cost

R. Tyler Croy rtyler at brokenco.de
Fri Feb 7 21:22:46 UTC 2020


We've been discussing some cost-mitigation measures with the Jenkins board
recently, and one of the big areas of cost for our infra is ci.jenkins.io.

An idea that struck me yesterday was that we could just spend a few thousand
dollars on a couple intel NUCs, keep those as "always on" agents, and then
use AWS/Azure/GCP for burst capacity on top of that.

In my home lab, I already have a cabinet with a few machines, cheap solar
power, and a gigabit connection to the public internet. It would be relatively
little effort for me to provision a few NUCs on their own network segment here.

The downsides that come to mind would be: physical infrastructure, a dependency
on Tyler's Great Little Datacenter, some personal cost to me (another network
switch and perhaps some electrical work to ensure the circuit will support even more
computers), and each machine would not have its own public IPv4 (though they
would all have public IPv6s).

The upside is that we've have basically some fixed cost executors, and even if
they fail, service availability would not completely disappear since we can
still burst into the cloud.


In terms of space, I believe I could fit up to 8 more NUCs in the cabinet. The
one I have right now has 8 cores and 32GB of RAM, which I think can easily
handle a couple core builds ;)




Let me know what you all think!

--
GitHub:  https://github.com/rtyler

GPG Key ID: 0F2298A980EE31ACCA0A7825E5C92681BEF6CEA2
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 839 bytes
Desc: OpenPGP digital signature
URL: <http://lists.jenkins-ci.org/pipermail/jenkins-infra/attachments/20200207/65abac68/attachment.asc>


More information about the Jenkins-infra mailing list