No announcement yet.

OSFMount: exFat or NTFS RAM Drive speed

  • Filter
  • Time
  • Show
Clear All
new posts

  • OSFMount: exFat or NTFS RAM Drive speed

    I want to mount an 8Gb ram disk
    -t vm
    Which will run 'faster': ntfs or exfat ?

  • #2
    What kind of disk access are you doing? Long linear reads / writes, or large numbers of small files and random access?
    How many threads hitting the RAM disk at the same time?

    My guess is exfat would be slightly faster sometimes, but for linear reads & writes I suspect there is much difference. It is easy to test however.


    • #3
      So I did a quick RAM Drive benchmark test to compare file system formatting options.

      Ryzen 5 5600x
      32GB DDR4 (4 x 8GB G Skill F4-3600C16)

      Settings in OSFMount for the RAM Drive:
      8GB Drive, GPT partition, exFat format and then NTFS format. Mounted as a volume (not a physical drive)
      OSFMount version 3.0 build 1006 to create the RAM drive.

      Benchmark Test:
      2 runs of PerformanceTest standard disk test, taking the best results.
      Note that these tests are single threaded. RAM Disks should do better with multi-threaded.

      Sequential read (MB/Sec) Sequential write (MB/Sec) IOPS 32KQD20 (MB/Sec) IOPS 4KQD1 (MB/Sec)
      exFAT RAM Drive 6801 4868 5471 1154
      NTFS RAM Drive 6122 4569 5091 1020
      SSD (Western Digital Black 500G M2 PCIe 4.0) 4124 3667 2993 101
      So on this quick test it looks like exFAT is around 5 to 10% faster than NTFS.
      And the RAM drives are between 50% and 1000% faster than a high end PCIe4 M2 SSD.

      4KQD1 = 4KBytes of data being read and written to the drive at a queue depth of 1
      32KQD20 = 32KBytes of data being read and written to the drive at a queue depth of 20

      Example screen shot from PerformanceTest
      Click image for larger version  Name:	exFat-RAMDrive.png Views:	0 Size:	62.8 KB ID:	49969


      • #4
        Many thanks.

        Mounted as a volume (not a physical drive)
        I only want to use "OSFMount CLI" to mount 'Empty RAM drive'. 'Disk Image file' not needed.

        Start-Process "D:\Customize\Win10\Install\OSFMount\osfmount. com" -Args "-a -t vm -m E: -o format:exfat:'RAMD' -s 8G" -Wait

        As a result, we get a physical driver?

        How do I mount a logical volume?

        Don't see in osfmount_Help (CLI).


        • #5
          OSFMount can either create a RAM drive that emulates a disk volume or an emulates physical hard drive.

          Disk volumes are more like partitions or shared network drives or cloud drives. You get a drive letter, but no knowledge of the physical storage medium.

          Physical drive emulation, as per the name, emulates a physical drive. You can partition it into multiple volumes, read low level sectors and request physical information about the drive (e.g. geometry, number of tracks, number of sectors, etc..). It will also appear in the Window disk management window and the drive would need partitioning before being useful.

          Volumes tend to have less overheads that emulated physical drives.

          The command line above will create a RAM drive as a volume and assign it to Drive E:. So think it does what you want.

          For the record, here is the DOS command line command to create a 8GB physical drive.
          osfmount -a -o physical -t vm -s 8G

          Here is a 2nd example that creates a new 1GB drive physical, partitions with GPT, formats it with exFAT, and adds a drive letter.
          osfmount -a -o physical -t vm -s 1G -o format:exfat:"RAMDISK" -o gpt


          • #6
            Thanks, basically understandable.


            • #7
              Just for comparison I re-did the benchmark with physical drive emulation (rather than volume emulation) with exFAT. Everything else was the same.
              Sequential read (MB/Sec) Sequential write (MB/Sec) IOPS 32KQD20 (MB/Sec) IOPS 4KQD1 (MB/Sec)
              3244 2856 2984 578
              As expected there is significant extra overhead emulating a physical drive (more layers of software, so higher latency). It is barely better than a high end SSD for single thread testing.