The YCSB 100 update and 128 thread load workloads generate a lot of dirty content in cache, writing that back to SLS efficiently is critical to the performance of the workloads.

Observing what content is being written back, how efficient those writes are (e.g: should they use deltas or not), and whether those writes are later being read back from SLS is critical to being able tune eviction in a way that effectively uses CPU (for reconciliation) and I/O bandwidth (for eviction and read-back).

A critical part of this will be observability into delta generation for writing back changed documents on pages efficiently. The YCSB 128 thread load relies on delta pages to efficiently track index creation - which is loaded in random order.

The outcome of this ticket would be a view or set of views in t2 that capture reconciliation/eviction/delta generation behavior, where we can observe changes in those metrics leading to improved throughput for the workloads being tracked.

This should be done in reference to the linked dependency tickets, that suggest possible improvements to cases when deltas are being generated.