Hi, I've run a series of benchmarks to compare PostgreSQL performance with various file systems and block sizes (file system as well as database). Nilfs2 was included in the benchmark and I was surprised how well it performed. I do have a few questions, though. All the following charts are with 4kB fs blocks and 4kB db blocks and a blue area means a checkpoint (i.e. a write activity). I've used kernel 2.6.39-gentoo-r3 and nilfs-utils-2.0.21. 1) write barriers With all the other file systems, the performance is significantly influenced by the write barriers, but this is not true for nilfs2. This is the tps I get with write barriers enabled: http://www.fuzzy.cz/tmp/nilfs2-barrier-4096-4-tps-rw.png and this is with write barriers disabled: http://www.fuzzy.cz/tmp/nilfs2-nobarrier-4096-4-tps-rw.png Both charts look exactly the same - that's a bit suspicious. For example with btrfs, the results are these http://www.fuzzy.cz/tmp/btrfs-datacow-barrier-4096-4-tps-rw.png http://www.fuzzy.cz/tmp/btrfs-datacow-nobarrier-4096-4-tps-rw.png so there's a significant difference. 2) continuous decrease of peformance One thing I don't like is that the performance continuously decreases - this is visible on the charts above. I do understand that the performance decreases during the checkpoint, but I'd expect it to grow after the checkpoint (and that does not happen). So I'm wondering - isn't there a mount option that could improve this? Tomas PS: The complete benchmark results are here http://www.fuzzy.cz/bench/. -- To unsubscribe from this list: send the line "unsubscribe linux-nilfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
