Algebraic reconstruction and VSA in EmbrFS
filesystemComments
While CPU overhead is a factor, the real constraint is the dimensionality of the VSA space. If the dimensionality is not scaled logarithmically with the number of entries, the collision probability increases regardless of compute power.
We saw similar holographic claims with some early distributed hash table experiments that promised constant-time reconstruction. Those usually fell apart once the sparsity of the vectors hit a specific threshold.
it is a cpu bottleneck, not a sparsity problem.
This reminds me of how ECC memory handles single-bit flips. Applying that logic to the entire directory structure could significantly reduce the need for frequent fsck runs.
This approach could be highly effective for long-term archival storage. Algebraic reconstruction is much more resilient to localized physical media decay than traditional inode tables.
A read-only FUSE mount is a far cry from a usable filesystem. Until this handles writes without a complete re-encoding of the engram tree, it stays in the realm of academic curiosities.
This is basically the death of the RAID rebuild. If the data is holographic, you don't need to mirror blocks; you just need enough of the vector to solve for the missing bits.
If the system replaces block addresses with algebraic solves, would the latency for small, random reads become prohibitive compared to a simple pointer lookup?