Whilst pursuing reports of an unresponsive/slow Dali, the extended Dali tracing showed that graph progress was being fetched very frequently.
For one particular workunit, there were 415K graph progress requests in a 6 minute period.
These corresponded to 53 WUInfo requests (logged in esp).
This workunit had a lot of subgraphs, and each request was retrieving elements data for 1 subgraph.
It's not obvious that this interaction was the direct or only cause of the slowdown, but it a lot of traffic, that could be done more efficiently.