I've got a problem since I switched to Linux some time back. Write (and read I guess) access to my HDDs is somewhat slow. I am organizing my data new and the copy time is killing me. I had this issue with different OS (Linux Mint 13, Debian Testing, Arch Linux). This issue occurses on several different kind of HDDs ranging from 500 GB to 3 TB, all connected with SATA (IDE and AHCI mode). The speed varies with the file system used but even with ext4 it's not really fast. If I test the (read) speed with some system tools (e.g. fdisk, gnome partition tool, ...) the speed is fine (> 100 MB/s). Copy speed also lacks when using command line tools e.g. rsync.
NTFS: ~20 MB/s
ext4: ~40-50 MB/s
btrfs: ~30 MB/s
Those benchmarks apply to copies on the same disk and from disk to disk as well. I am using a AMD Phenom II X955 Processor on an AsRock 890GX Pro 3 Mainboard.
Does anyone of you encounter such an issue? Can you tell me the speed one can/should reach with NTFS? I would like to set up at least one NTFS partition (as I run a small Windows setup on the same machine) but with write speed of about 20 MB/s this isn't really usable.
This depends on various things.
1.) Using IDE emulation or native AHCI makes a huge difference. You should always be using AHCI.
2.) Are your disks SATA 2 or 3? Are those 5400rpm "eco" disks or 7200rpm disks, and what's their cache size?
3.) What type of files are you copying? Lots of small files, or huge files?
4.) Depending on the FS used and the overall workload of your box, choosing a different I/O scheduler could make some difference. Deadline is more oriented towards straight throughput, whereas the default cfq scheduler does fair queueing so not to block overall interactivity.
5.) For EXT4 and BTRFS you should be mounting using "-o noatime" or "-o relatime"
I have no idea about NTFS. Is that measured from inside windows, or did you use the ntfs-3g driver?
What kind of sequential write speed do you get on Windows on this very drive? Use files 2x times bigger than your RAM size to avoid caching bias.