AFP transfer sometimes very(!) slow

  • And sleeping every frickin 7 seconds is so life preserving ;)


    Greetings
    David

    "Well... lately this forum has become support for everything except omv" [...] "And is like someone is banning Google from their browsers"


    Only two things are infinite, the universe and human stupidity, and I'm not sure about the former.

    Upload Logfile via WebGUI/CLI
    #openmediavault on freenode IRC | German & English | GMT+1
    Absolutely no Support via PM!

    Edited once, last by davidh2k ().

  • For the price it deserve a WD-RE or an HGST UltraStar. Same power savings, better reliability.

    HP N54 MicroLiant - 4x4To WD RED RAID 5 - OMV 0.5.53


  • You have to take those settings with a grain of salt. You only want to adjust that for writing performance issues, should not affect reads from the NAS. For my setup, 256 pages was working okay but it wasn't best. I reduced it to 128 pages and I was able to get constant 120MB/s writes. That's with my current setup of x3 drives on RAID5. For a friend of mine, the 256 pages cache for stripe was best, and he has RAID6 with about 12 drives. Going under or above has serious side effects on the writing performance, I don't understand how people can adjust that at such an insane high level and expect good results. There's a reason for the Debian devs to be conservative about this setting.

    Absolutely no support through PMs!!
    I work for the HPE R&D division. Have tech questions about ProLiant servers? Ask me!
    Learn more about HPE's iLO included in all ProLiant servers.

  • You have to take those settings with a grain of salt. You only want to adjust that for writing performance issues, should not affect reads from the NAS. For my setup, 256 pages was working okay but it wasn't best. I reduced it to 128 pages and I was able to get constant 120MB/s writes. That's with my current setup of x3 drives on RAID5. For a friend of mine, the 256 pages cache for stripe was best, and he has RAID6 with about 12 drives. Going under or above has serious side effects on the writing performance, I don't understand how people can adjust that at such an insane high level and expect good results. There's a reason for the Debian devs to be conservative about this setting.


    Yeah already tried adjusting striping (128k or 64k gives best R/W), and it works well on SMB, but nothing on AFP. But as I said earlier, I'm happy with SMB. I keep AFP for TimeMachine only and it's better that way. I'm only waiting for AFP being executed from Apple, and then I'll throw a big party.

    HP N54 MicroLiant - 4x4To WD RED RAID 5 - OMV 0.5.53

  • Well I sort of steered away from the subject, sorry about that. It is odd that you have that problem though. I do have mine with AFP and I am getting full 120MB/s speeds and I didn't have to do anything special to achieve that. My only suggestion is to turn off Service Discovery from the Network section on OMV for everything but AFP. Also make sure that the name represented in that section is the same you use in the description section for SMB. That way, OSX will know is the same server and only present you with the AFP mount options. My only theory is that you may be using a service that may not actually be AFP, even if it works. I had a hard time telling the services apart until I turned off discovery on everything but AFP.

    Absolutely no support through PMs!!
    I work for the HPE R&D division. Have tech questions about ProLiant servers? Ask me!
    Learn more about HPE's iLO included in all ProLiant servers.

  • Yeah I tried that too. I accessed the share using adress bar with afp://.


    But in fact it's a software isssue on 10.9. With a VM of OSX 10.8 on my linux laptop, about 95Mbps R/W. Mystery solved.


    I'll try to install hackintosh 10.9 on my laptop for further testing.

    HP N54 MicroLiant - 4x4To WD RED RAID 5 - OMV 0.5.53

  • Well I just noticed a pattern on your logs:



    And this:



    It is very odd, but it could be a Time Machine job gone wrong attempting again and again to finish or start. I've seen this kind of behavior before with Snow Leopard and even Lion. All I did was remove the old backups from the NAS and redo them from scratch. Let me know how that goes.

    Absolutely no support through PMs!!
    I work for the HPE R&D division. Have tech questions about ProLiant servers? Ask me!
    Learn more about HPE's iLO included in all ProLiant servers.

  • Yeah I tried that too. I accessed the share using adress bar with afp://.


    But in fact it's a software isssue on 10.9. With a VM of OSX 10.8 on my linux laptop, about 95Mbps R/W. Mystery solved.


    I'll try to install hackintosh 10.9 on my laptop for further testing.


    So as I suspected, AFP <-> Netatalk is pretty shitty in 10.9


    OSX 10.9 : +- 35Mbps
    Hackintosh 10.9 : +- 35 Mbps
    Hackintosh 10.8 : +- 95 Mbps
    VM 10.8 on Debian : +- 95Mbps


    Maybe a netatalk compatibility issue. Anyway I never doubted OMV :)

    HP N54 MicroLiant - 4x4To WD RED RAID 5 - OMV 0.5.53


  • debian/OMV uses Netatalk 2.2.2 which is 2.5 years old version. If you give me your testing methology, I can perform tests on 2.2.2 and 3.1.0 for comparison.

  • The old netatalks have some features missing. At leaste the Netatalk version for Kralizec performs ok. But still i would prefer a actual version


    BlackMagic Speed Test: up to 120MB


    Finder: 2 concurrent copies checked via Activity Monitor/Network: 120MB/s ...


    XBench:


    Results 81.22
    System Info
    Xbench Version 1.3
    System Version 10.9.4 (13E28)
    Physical RAM 16384 MB
    Model MacBookPro11,3
    Disk Test 81.22
    Sequential 48.41
    Uncached Write 23.73 14.57 MB/sec [4K blocks]
    Uncached Write 164.27 92.94 MB/sec [256K blocks]
    Uncached Read 34.55 10.11 MB/sec [4K blocks]
    Uncached Read 183.34 92.14 MB/sec [256K blocks]
    Random 252.12
    Uncached Write 129.83 13.74 MB/sec [4K blocks]
    Uncached Write 289.57 92.70 MB/sec [256K blocks]
    Uncached Read 370.97 2.63 MB/sec [4K blocks]
    Uncached Read 496.60 92.15 MB/sec [256K blocks]

    Tom


    ----


    HP N54L, 6GB, 5disc Raid5, SSD Boot with OMV 5
    HP N54L, 16GB, 4disc ZFS pool, SSD Boot with other NAS system


  • debian/OMV uses Netatalk 2.2.2 which is 2.5 years old version. If you give me your testing methology, I can perform tests on 2.2.2 and 3.1.0 for comparison.


    I transfered a 45Gb video, using both ultracopier and finder.

    HP N54 MicroLiant - 4x4To WD RED RAID 5 - OMV 0.5.53

  • Hmm, i transferred 4 TB today, OS X 10.9.4 via afp to omv (1.x). Transfer speed 80mb/s (the source drive wasn't faster). Mostly big files (100mb+)

    Tom


    ----


    HP N54L, 6GB, 5disc Raid5, SSD Boot with OMV 5
    HP N54L, 16GB, 4disc ZFS pool, SSD Boot with other NAS system

  • Hmm, i transferred 4 TB today, OS X 10.9.4 via afp to omv (1.x). Transfer speed 80mb/s (the source drive wasn't faster). Mostly big files (100mb+)


    I will try OMV 1.x then

    HP N54 MicroLiant - 4x4To WD RED RAID 5 - OMV 0.5.53


  • I transfered a 45Gb video, using both ultracopier and finder.


    I have cloned both vm's from fresh install of 1.0.14. One with included netatalk, other with compiled 3.1.0, just basic configuration.
    Both were between 105-119MB/s with one large file, only difference was OSX recognizing 3.1.0 as time capsule.

  • debian/OMV uses Netatalk 2.2.2 which is 2.5 years old version. If you give me your testing methology, I can perform tests on 2.2.2 and 3.1.0 for comparison.


    I was curious about netatalk 3, main difference is the shares can now be indexed by spotlight and of course a different syntax for configuration. How is the spotlight working for you?
    I've build a custom version for a VM now for testing and I am thinking of moving it to my main machine. I'll have to leave the omv netatalk interface behind.
    Also stock wheezy is 2.2.2. Version 2.2.5 is in the Sid repo, has the main difference that allows symlinks which I find very useful

  • I was curious about netatalk 3, main difference is the shares can now be indexed by spotlight and of course a different syntax for configuration. How is the spotlight working for you?
    I've build a custom version for a VM now for testing and I am thinking of moving it to my main machine. I'll have to leave the omv netatalk interface behind.
    Also stock wheezy is 2.2.2. Version 2.2.5 is in the Sid repo, has the main difference that allows symlinks which I find very useful


    I use AFP for timemachine only so I dont use spotlight there at all.

  • I was curious about netatalk 3, main difference is the shares can now be indexed by spotlight and of course a different syntax for configuration. How is the spotlight working for you?
    I've build a custom version for a VM now for testing and I am thinking of moving it to my main machine. I'll have to leave the omv netatalk interface behind.
    Also stock wheezy is 2.2.2. Version 2.2.5 is in the Sid repo, has the main difference that allows symlinks which I find very useful


    Im presently in the middle of an OMV3 build and thought it was plain sailing from here - but I can't get spotlight to index my AFP share. Is any special configuration required to get this to work? Thanks

    NAS OMV 3.0.59 Rig: DFI Lanparty nForce4 SLI-DR | Opteron 165 Dual Core (no overclock) | 4GB Corsair Non-ECC DDR400 RAM | OMV3: 1x100GB IDE | Data: 2x 250GB WD 7200rpm, 2x 3TB WD Red, 1x 4TB Samsung 5400rpm

  • Sorry to disappoint you but spotlight does not refer to spotlight OS X (took me a while to figure this out). Is a feature intended for gnome tracker (search engine for Linux desktop). If you want spotlight to index AFP I think you need to configure in OS X using cli so spotlight indexes the volume, this IMO it doesn't work. Maybe it works better with some net storage solutions from apple like airport.

Participate now!

Don’t have an account yet? Register yourself now and be a part of our community!