First thing: if you're writing to the device (like /dev/sdX), then you'll get sequential writes at the device. If that device is a spinning disk, then you'll get the best possible performance of laying down blocks one after another. If the device is a SSD, then they'll tend towards filling erase-block size chunks, but you won't have any idea of physical location. If you're writing to files on a filesystem, however, the answer is less straightforward. Most filesystems will attempt to translate sequential file writes to mostly-sequential writes to the underlying disk. There will be metadata updates in there and whatnot that shake things up a bit. If the filesystem is copy-on-write, like ZFS, and it's been used a lot (has lots of holes), then there's no telling. Finally, if the device is a RAID, SAN box, or whatever, then that can be very different than a raw disk device. Best regards, Josh On Jan 26, 2015, at 5:56 AM, telecolega . <telecolega@xxxxxxxxx> wrote: > Hi guys, > > In the last days I tried to investigate how fio writes sequentially > when using the libaio engine. As far as I understood, the key here is > controlling the offset variable in the iocb struct (iocb->u.c.offset) > so that the new block is written right next to the previous block. > > Now a question comes to my mind. Even if fio is writing sequentially > on a file or block, that does not guarantee that we are writing > sequentially on the hard disk, right? I guess the linux kernel will be > the one eventually deciding if that data is written sequentially or > randomly across the disk. Nevertheless, I assume that when FIO writes > sequentially to a file or block, the likelyhood that the data is > written sequentially is increased. Is this correct? > > Do you know if there is a way from the user space (e.g. fio) to > "force" the kernel to write sequentially? > > Thanks ofr your help! > -- > To unsubscribe from this list: send the line "unsubscribe fio" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe fio" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
