4 ms·

On the related topic of store buffers and write coalescing, can anyone point me to some good articles discussing how this works in practice for merging small wr

On the related topic of store buffers and write coalescing, can anyone point me to some good articles discussing how this works in practice for merging small writes to the same cache line? I have a workload where we are appending different size payloads (of sizes between 1 and 64 bytes) to an in memory log. One option is to just store each entry in its own cache line. A more space efficient option is to pack each write tightly one after the other. My hypothesis is that due to write coalescing in store buffers this will also be more memory bandwidth efficient since writes to the same cache line will be merged. Is this correct? Note that metadata for each entry (e.g. the size) will be stored separately.

gpderetta 3y ago
On a typical write back cache, the cache line is written back to memory (or to a lower hierarchy) only when needed: either some other core wants it or it needs to be evicted to make space for other data. For a write mostly task, if you write to the same line again, it is likely to be still in the M state and data can be written to it immediately. Later, when the line is written back, the separate writes are effectively coalesced. But if some other core is concurrently reading from the head of the log, it can be suboptimal as the cacheline will keep bouncing between M and S state generating coherence traffic. This is still more scalable than two threads writing to the same cacheline of course.