>>>>> "Aleksander" == Aleksander Adamowski <linux@xxxxxxxxxx> writes: Aleksander> I did some extensive experiments and benchmarks, and the Aleksander> conclusion is that the drive actually uses 4 kB sectors Aleksander> internally, although it doesn't report that fact to the Aleksander> outside world. That's really lame! You should ask for a firmware update. For a while now I've been lobbying for SMART metrics we could query to get an idea of aligned vs. misaligned requests submitted to a drive. That would be handy in a situation like this. Unfortunately the storage industry is glacially slow. As witnessed by the fact that the 4KB sector transition was supposed to be complete for the release of Windows Vista. And now we're talking 2012 or so... Aleksander> These results clearly indicate that the drive has a sweet Aleksander> spot with partition starts aligned to sectors divisible by Aleksander> 8: performance on partitions starting at sectors 40, 48, 56, Aleksander> 64 is roughly 5.5 times better that on all others. Aleksander> This is a bit puzzling - 5.5 x faster is more that one would Aleksander> expect from a simple read-modify-write issue, isn't it? I've Aleksander> read about performance differences of 2:1, not 5.5:1. It really depends on the size of the I/O. You pay a penalty at the first and last partial 4KB block. The penalty is caused by rotational latency. The drive first has to seek to read the block and then wait for the same spot to come back under the head so it can write it. IOW, the penalty depends highly on the rpm. I assume they read ahead the runt so the effect of that may be less pronounced. Aleksander> So for any other owners of WD EARS drives, if these don't Aleksander> report physical 4096-byte sectors to you, don't believe them Aleksander> and align your partitions at the aforementioned sectors (a Aleksander> generally good idea is to run the postmark benchark to Aleksander> compare performance on aligned and non-aligned partitions). Just last week we were discussing aligning everything on a 4KB boundary by default. The fact that you have a rogue production drive is disconcerting but it helps emphasize the value of a new default. -- Martin K. Petersen Oracle Linux Engineering -- To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
