ref: de88b79d230497866fe0c6afbefff5d7f283d46a
parent: 856c1dd4e268797581995d9addc0aa3c939e100a
author: Ori Bernstein <ori@eigenstate.org>
date: Fri Apr 28 11:20:08 EDT 2023
gefs.ms: we did that one.
--- a/gefs.ms
+++ b/gefs.ms
@@ -679,10 +679,6 @@
Having native RAID would also allow automatic recovery based on block pointer hashes.
This is not possible with fs(3), because if one copy of a block is corrupt, fs(3) would not know which one had the incorrect data.
.PP
-Furthermore, growing the file system would be extremely useful for taking flashed disk images and growing them to the full size of the underlying storage.
-The way that arenas work allows for this to happen fairly easily, but right now there's no way to change the list of arenas.
-Additionally, the naive round robin allocation across arenas works doesn't allow for balancing.
-.PP
There are a number of disk-format changing optimizations that should be done.
For example, small writes should be done via blind writes to the tree. This would allow small writes to be blindingly fast,
and give us the ability to keep small files directly in the values of the tree, without allocating a disk block for them.