Idle discussion - multi-target workflows

[bjackman]

Hi folks, I have been spending lots of time lazily pondering systems like WA lately, considering their relative advantages and disadvantages. I think the only significant disadvantage I could think of for WA (aside from silly things like "doesn't have workload X" and "I wish it wasn't Python"!) is that AFAIK it's not really possible to run client/server type workloads (right?).

I spent a very short few minutes looking at the code, and it doesn't actually seem like this limitation is particularly fundamental. I'm curious if there's anything I'm unaware of that would make it very tricky to implement? (Or indeed if there's something I'm unaware of that means it's actually already possible)? Like basically could there be a variant of the Workload class that has targets instead of target?

Some thoughts that come to mind: at present WA has good capabilities to inform the workload about, or abstract away, information about targets. But in a multi-target world you probably also care about relationships between targets for example:

• if/how they can communicate over the network
• whether this network is a stable wired LAN/wireless network/over the internet
• if one device is a USB gadget attached to another device
• if the devices are in physical proximity to one another (e.g. I have used a WA-alike for benchmarking WiFi drivers, in that case we would have wanted to avoid accidentally benchmarking bandwidth between devices in different faraday cages! This is slightly different from "are the devices connected by the network" because the network connection actually gets created as part of the workload)

Would that be over-thinking it? Any other thoughts?