Potential issue with mapping of block_id to object in Meta Tectonic OracleGeneral format

[shermanjlim]

Description

In the OracleGeneral files for the Meta Tectonic traces, it seems like each object corresponds to a 'block' in the original Meta traces.

However, in Meta's own simulator for the same storage traces, they break down each block into chunks instead, and each chunk is the unit of a cache object.
Reference: facebook/CacheLib@23a888e#diff-c2dedb4daf415ffac8169c4633a4790fd252172d3b45b04c96577d5d70a76d26
"if the chunk size is 256KB, the full extent of the block of size 8MB will be cached as 32 cache items"

[1a1a11a]

Thank you for bringing this up! Yes, the few storage dataset are not properly converted (not respecting the original semantics), are you interested in leading an effort converting the traces?

[shermanjlim]

Thanks for confirming! I plan to use libCacheSim to run some simulations with Meta's storage traces, so I'll try to convert those correctly. Will definitely share what I have after I'm done!

[1a1a11a]

Awesome!