Factors that affect speed are: /dev/sd*
* are cached (cache has a 3GB cap presently since the DMA pushback diff not was experienced as stable and therefore rolled back), which may make access appear faster * I think the underlying hardware access is always split to 512B (or in certain cases 4096B) accesses by the file/IO subsystem, this together with serialization via the kernel biglock and subsystem design gives you a system-global cap of about 120MB/sec independent of actual hardware. /dev/rsd* is * not cached * i think underlying hardware access here depends on the size of the read operations, so for dd(1) this would be the size of the "bs" argument. For normal SSD:s the internal block size of 16KB (or any even multiple of 16KB) would give you the highest performance, for an SSD this would be about 500MB/sec in both sequential and random access. I see as the *highest priority* in OpenBSD to make the disk/file IO subsystem parallelized all down to the point of using the hardware's multiqueing feature. Currently, parallel disk/file accesses in software are all serialized. Modern SSD:s all give you about 80MB/sec access in normal sequential read mode. As you parallelize it you will see an linear speed increase with higher number of threads up to approx 10 threads, where a SATA SSD will give you about 500MB/sec and a PCIe NVMe SSD will give you about 900MB/sec. The last file/disk IO related matter to remind of is that mmap() IO not can be combined with file IO presently. On June 15, 2018 9:10 AM, Maximilian Pichler <maxim.pich...@gmail.com> wrote: > I'm getting much faster read speeds from an SD card when mounting the > card and reading files (~50MB/s) than when reading the raw device > rsd1c (~25MB/s). If anything, shouldn't it be the other way round, > given that the file system has some overhead? > > Here are the measurements when mounting and reading files: > > $ doas mount /dev/sd1i /card > $ mount | grep card > /dev/sd1i on /card type msdos (local) > $ cat /card/DCIM/100MSDCF/* | pv -bra -Ss500m | sha1 > 500MiB [49.2MiB/s] [49.2MiB/s] > e823304263649ee4aff5c4563878cd9111cffebf > > For the raw device (rsd1c): > > $ doas cat /dev/rsd1c | pv -bra -Ss500m | sha1 > 500MiB [25.5MiB/s] [25.5MiB/s] > ca17bdb9a657bbcf654a60057861be8fe02df0b1 > > For what it's worth, the block device (sd1c): > > $ doas cat /dev/sd1c | pv -bra -Ss500m | sha1 > 500MiB [8.53MiB/s] [8.53MiB/s] > ca17bdb9a657bbcf654a60057861be8fe02df0b1 > > The card was previously formatted with "newfs_msdos -s 8000000 sd1c" > and then files were written on it by a camera. > > Here is the card reader's dmesg bit: > > umass0 at uhub0 port 13 configuration 1 interface 0 "Generic USB3.0 > Card Reader" rev 3.00/15.32 addr 5 > umass0: using SCSI over Bulk-Only > scsibus2 at umass0: 2 targets, initiator 0 > sd1 at scsibus2 targ 1 lun 0: <Generic, STORAGE DEVICE, 1532> SCSI4 > 0/direct removable > sd1: 488154MB, 512 bytes/sector, 999740416 sectors > sd2 at scsibus2 targ 1 lun 1: <Generic, STORAGE DEVICE, 1532> SCSI4 > 0/direct removable
