I'm trying out bcache with a new use case.
I have a cache volume of 1350G and a backing raid array of 15TB. The
cache volume is a normal hard disk, not an SSD. The goal is to prevent
accesses to the raid array, allowing it to be kept in a spun down state.
My tests show that this indeed works.
I've set up the cache as write-through, disabled sequential cutoff and
disabled read/write congestion by setting all of those to 0.
Here are some statistics:
--- bcache ---
Device /dev/bcache0 (251:0)
UUID 4cc40843-962c-43b3-9fa2-a3b88f56c3d1
Block Size 0.50KiB
Bucket Size 512.00KiB
Congested? False
Read Congestion 0.0ms
Write Congestion 0.0ms
Total Cache Size 1.32TiB
Total Cache Used 256.50GiB (19%)
Total Cache Unused 1.07TiB (81%)
Dirty Data 0B (0%)
Evictable Cache 1.32TiB (100%)
Replacement Policy [lru] fifo random
Cache Mode [writethrough] writeback writearound none
Total Hits 123136 (31%)
Total Misses 264135
Total Bypass Hits 2929 (100%)
Total Bypass Misses 0
Total Bypassed 11.40MiB (<-- before cutoff was set 0)
In the process of loading the cache with data, I've been accessing a few
files (>20G+) to get them cached by hashing them with md5sum. This
works well when readahead=0. The cache misses go up, total cache used
goes up, etc. exactly what you would expect. Resumming them again later
results in all cache hits --> excellent.
However, when I set readahead to 1.0M, suddenly bcache is claiming that
all reads I'm doing (of files I'm 100% sure are not in the cache yet)
are cache *hits*, and total cache used stays the same. However, even
though bcache claims all reads are hits, I can see the backing device
getting accessed... hardly hits then? What's worse, it is not adding
the data to the cache anymore.
As soon as I disable the readahead again by setting it to 0 (without
even stopping the md5sum process), cache misses go up again and the
total cache used value goes up as well.
Perhaps I'm misunderstanding what readahead does? Or perhaps it is
acting differently when the cache device is not an SSD?
--John
--
To unsubscribe from this list: send the line "unsubscribe linux-bcache" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
