My educated speculation is that builds are particularly sensitive to I/O latency (read this file, and based on its contents, read this other file). For that kind of workload, the directly attached NVME that you get in a laptop will beat the pants off the network attached disk that you get in a budget friendly cloud instance.
This is absolutely true.
We run fairly heavy duty CI infra and I can say with confidence that this is one of the biggest factors impacting CI runner performance, almost equal in weight to raw CPU perf.
I think it's also fair to rope in how fucking bloated so much software is now. The size of many applications that do garden variety work like real time chat is fucking ridiculous. Slack at my last time I looked was just under 200 MB. TWO. HUNDRED. MEGS. To send and receive fucking instant messages and do voice/video calling.
Comments
My educated speculation is that builds are particularly sensitive to I/O latency (read this file, and based on its contents, read this other file). For that kind of workload, the directly attached NVME that you get in a laptop will beat the pants off the network attached disk that you get in a budget friendly cloud instance.
This is absolutely true. We run fairly heavy duty CI infra and I can say with confidence that this is one of the biggest factors impacting CI runner performance, almost equal in weight to raw CPU perf.
I think it's also fair to rope in how fucking bloated so much software is now. The size of many applications that do garden variety work like real time chat is fucking ridiculous. Slack at my last time I looked was just under 200 MB. TWO. HUNDRED. MEGS. To send and receive fucking instant messages and do voice/video calling.