Menu

A few bugs/question (subnets, kernels, speed)

Anonymous
2009-08-29
2013-04-05
  • Anonymous

    Anonymous - 2009-08-29

    Hello, we use clonezilla server to clone 30 machines each machine is about 500GB in size.

    I had a few problems after updating to the latest jaunty.  The network card in our client machines didn't want to work with the latest kernel so I had to compile my own with the drivers built into the kernel.  This presented some problems with drblsrv as it would not allow me to use a custom kernel and instead auto downloaded the latest kernel.

    After some tinkering I got it to accept my custom kernel, would it be possible to allow the option of a custom kernel in DRBL?

    It seems that if the server is on a /8 subnet (netmask 255.0.0.0) it will ignore the client names in /opt/drbl/conf/client-ip-hostname, if the subnet is a /24 (netmask 255.255.255.0) everything runs as expected.

    Also I was hoping you could give me some tips on speed, unicast will run very fast when saving and pushing an image but multicast runs very slow. 

    Example:
    Unicast
    Client 10.0.2.108 (00:19:d1:93:45:a0) finished cloning. Stats: Unicast restored 2009-08-27-18-img, /dev/sda1, success, 517735 MB, 91.398 mins, 5664.0 MB/min;

    Multicast
    Client 10.0.2.108 (00:19:d1:93:45:a0) finished cloning. Stats: Multicast restored 2009-08-27-18-img, /dev/sda1, success, 517735 MB, 564.374 mins, 912.0 MB/min;

    Any help appreciated.

     
    • Jorge Martinez

      Jorge Martinez - 2009-08-30

      I have the same problem about speed modes, unicast seems to be faster than multicast, if you get to know something about how to improve multicast speed please let me know

      regards

       
    • Steven Shiau

      Steven Shiau - 2009-08-31

      Yes, DRBL now only supports clients with netmask 24 (255.255.255.0), thanks, we will improve that.
      For customized kernel, you can try the offline mode:
      http://drbl.sourceforge.net/one4all/#drblsrv-offline

      As for the multicast speed, maybe you can refer to this:
      http://drbl.sourceforge.net/faq/fine-print.php?path=./2_System/67_multicast_slow.faq#67_multicast_slow.faq
      It's possible to tune udpcast parameters to make that.

      Steven.

       
      • Jean-Philippe Menil

        Hi,
        If order to use multicast, you must configure your switch properly to use it.
        Unicast seem to be faster, because it broadcast the paquets on all the ports of the switch.
        Multicast is better, because all the client register to a group, and the paquest are forwarding to this group only.

        Exemple:
        You have one switch of 24 ports, 12ports with the machine you want to restore, 12 other ports for a classe of student
        In unicast mode, the speed of the "restore" is good, but the student have no more network due to "broadcast tempest".
        In multicast mode, the network speed is good too, and your student can still work, because they don't receive the "restore flux".

        What is your switch?

         
        • isrc

          isrc - 2009-08-31

          HI Jean-Philippe,
          As you seems like our multicast expert, would you mind sharing the multicast configuration of  your(s) switch(es)  please ?
          I've some HP procurve and Cisco 29x switch (tried enabling img snooping) lying around , and i must confess that how to get multicast working as the stream ends up always being broadcasted on all ports or blocked !?
          I've found some post talking about a possible issue with igmp v3 vs igmp v2 but not sure what it's all about

          Thanks !

           
          • Jean-Philippe Menil

            Hi,
            i'm not a multicast expert ;)
            On your cisco:
            To enable IGMP snooping on the switch, issue the ip igmp snooping command.
            To enable IGMP snooping on a VLAN interface, issue the ip igmp snooping vlan command.
            If global snooping is disabled, VLAN snooping cannot be enabled.
            On your hp:
            vlan 1 (or/and vlan xxx, vlan 1 is the default vlan)
            ip igmp

            See this post for the tcpdump command to verify if multicast work or not.
            What's your kernel version on your server?

             
            • Jean-Philippe Menil

              Arf,
              i've forget the link of the post:
              https://sourceforge.net/forum/message.php?msg_id=7383498

               
              • isrc

                isrc - 2009-09-01

                Hi,
                Thanks  Jean-Philippe for taking the time to give some detail ! Will try those commands  this week and report back here any success :-) Do you think that a cisco 2960 or hp  2124 could be ok for those tests ?

                I'm running DRBL drbl-1.9.4-27.i386 on Fedora 10 with a up to date 2.6.27.25.. The kernel booting the clients is (aka /tftpboot/nbi_img/vmlinuz-pxe)  is vmlinuz-2.6.27.21-170.2.56.fc10.i686

                Thanks again for your help.

                 
  • Anonymous

    Anonymous - 2009-09-16

    The switch we are using is a procurve 2848, with IGMP snooping enabled on the VLAN we are using.

     
  • Anonymous

    Anonymous - 2009-09-16

    Alright it's all sorted!  Now getting 4300MB/min on a multicast!  Should go up a little bit when I move to a faster RAID array.

    The problem was UBUNTU, I tried both desktop and server but both had the exact same problem of the multicast starting fast but quickly slowing down to around 1500MB/min.  I have installed Debian desktop edition and the speed is a lot better and more stable.

     
  • Steven Shiau

    Steven Shiau - 2009-09-16

    So maybe the kernel from Ubuntu does not support your hardware very well… But Debian does.
    Thanks for sharing that.

    Steven.

     
  • Anonymous

    Anonymous - 2009-10-08

    Steven I am not sure if it is hardware support or a different issue, running the server on a different motherboard/CPU/RAM/HDD/RAID produced identical results.  Slow on Ubuntu and fast on Debian, the strangest thing was that the speeds were almost the exact same on different hardware platforms.

    After a bit of time it looks like the server is getting faster, multicasting has moved to about 5300MB/min

     
  • Steven Shiau

    Steven Shiau - 2009-10-12

    mrmacomouto,

    I guess it's kernel/hardware support issue.

    Steven.

     

Log in to post a comment.

MongoDB Logo MongoDB