Slow TFTP downloads RRS feed

  • Question

  • Hi everyone,

    This one really has got me baffled.  Our Lite Touch image is ~200MB, and takes about 1 minute and 15 seconds to download from WDS if one client is booting the WIM.

    The moment I do two clients, it goes up to about 2 minutes (and so on and so forth).

    Regardless of this, it seems our WDS server has set a cap of 0.1% of the available bandwidth for TFTP.

    The server itself is an HP DL180G6 with 4 gigabit NICs all teamed in 802.3ad Link Aggregation using the HP teaming software.  The clients are mostly HP dx2450 machines with NVIDIA gigabit bootable NICs.

    I can't seem to find a way to persuade WDS that it can use more than 0.1% of the bandwidth.

    We have adjusted the TFTP block size, and any more than 1422 causes problems (we lack jumbo frame support in places!), and we've tried TFTP window sizes from 1 through 16 - no performance increases (many decreases at higher numbers!).

    We've even been through as well, and upped the buffers on each card to 2048 (although we can't seem to adjust them on the team itself in the HP software!)

    Has anyone got any ideas?  This is driving us up the wall!



    Friday, August 9, 2013 7:02 AM

All replies

  • Are you using multicasting by any chance? Also take a look at this

    I know its a bit old but maybe it will help in your case. You may also want to start investigating any network related settings, maybe on the switch for example, that is causing the issue.

    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread. ”

    Friday, August 9, 2013 12:09 PM
  • We're on Server 2008 R2 here - so the Network Profile thing goes out the window.

    Multicast is also out of it, mainly because we can't get it work properly on a routed network - so it's off.

    I don't think it's switch related - I'll have to build a dev environment somewhere to test that though - we've exhausted the config options on the switches as far as we can see - so it'll be a case of getting a dumb gig switch involved and testing there.

    I'm wondering if it's related to our LACP trunks... but not sure.

    Our WDS server sees high utilisation on one core (and one core only), and it seems to be related to network activity - so I am suspecting Kaspersky's filter driver may be involved.....

    I'll try and build a vanilla WDS/MDT setup and see how that performs.......

    Friday, August 9, 2013 12:15 PM
  • If I remember right, the boot image is always unicast. Only the install image can be multicast. 

    I'd be interested in the results from a vanilla WDS/MDT build. My boot image is about ~350 MB and takes, maybe, 30 seconds to download.

    -Nick O.

    Friday, August 9, 2013 1:55 PM