[Lustre-discuss] Lustre over two TCP interfaces

Alfonso Pardo alfonso.pardo at ciemat.es
Wed Jun 26 23:25:00 PDT 2013

Yes I have two swtiches, one to the bond0 interface and other switch to the second bond1 interface.

From: Indivar Nair 
Sent: Wednesday, June 26, 2013 8:05 PM
To: Alfonso Pardo 
Cc: Michael Shuey ; WC-Discuss ; lustre-discuss 
Subject: Re: [Lustre-discuss] Lustre over two TCP interfaces

Hi Alfonso,

I guess, you have two switches, with 2 interfaces (bond0) connected to one switch and the other 2 interfaces (bond1) to the second switch.


What you need to do is merge the switches using a 'stacking' cable (if the switches are stackable) and create a single trunk using 2 ports from each switch.
Then create a single bond on the Linux side using all the 4 Interfaces (and have just 1 IP).

Use bonding mode balance-rr or 0 without LACP to get load balancing across all the 4 NICs.

If the switches aren't stackable and a single trunk cannot be created on the switch side, then use bonding mode balance-alb or 6 on the Linux side.

No changes need to be done to the cabling in either case.


This way you get Load Balancing and H/A across NICs.

Indivar Nair

On Wed, Jun 26, 2013 at 5:47 PM, Michael Shuey <shuey at purdue.edu> wrote:

  That will probably be slow - the machine you use to proxy the IPVS address would be a bottleneck.  Out of curiosity, what problem are you trying to solve here?  Do you anticipate whole-subnet outages to be an issue (and if so, why)?

  Mike Shuey

  On Wed, Jun 26, 2013 at 4:53 AM, Alfonso Pardo <alfonso.pardo at ciemat.es> wrote:


    Thanks for you reply! May be another way is a floating IP between two interfaces with IPVS (corosync).

    -----Mensaje original----- From: Brian O'Connor
    Sent: Wednesday, June 26, 2013 10:15 AM
    To: Alfonso Pardo
    Cc: 'Michael Shuey' ; 'WC-Discuss' ; lustre-discuss at lists.lustre.org
    Subject: Re: [Lustre-discuss] Lustre over two TCP interfaces 

    On 06/26/2013 04:16 PM, Alfonso Pardo wrote:

      But.... if I configure the OST assigning to the first interface of the
      OSS (bond0) and as failover OSS the second inteface of the OSS. If the
      bond0 network down, the client will try to connect to the failover, that
      is the second interface of the OSS.
      is it possible?

    I stand to be corrected, but no, I don't think so. As I understand it
    the failover code looks for a different server instance, rather than a
    different nid.



      *From:* Brian O'Connor <mailto:briano at sgi.com>
      *Sent:* Wednesday, June 26, 2013 1:09 AM
      *To:* 'Alfonso Pardo' <mailto:alfonso.pardo at ciemat.es> ; 'Michael Shuey'
      <mailto:shuey at purdue.edu>
      *Cc:* 'WC-Discuss' <mailto:WC-Discuss.Migration at intel.com> ;
      mailto:lustre-discuss at lists.lustre.org
      *Subject:* RE: [Lustre-discuss] Lustre over two TCP interfaces
      Unless something has changed in the new versions of lustre, I don't
      think lustre can do failover between nids on the same machine.

      It can choose the available nid at mount time, but if an active nid goes
      away after you are mounted then the client chooses the failover nid, and
      this must be on a different server.

      Check the archives for more discussion in this topic :)

      -----Original Message-----
      *From: *Alfonso Pardo [alfonso.pardo at ciemat.es
      <mailto:alfonso.pardo at ciemat.es>]
      *Sent: *Tuesday, June 25, 2013 07:23 AM Central Standard Time
      *To: *Michael Shuey
      *Cc: *WC-Discuss; lustre-discuss at lists.lustre.org
      *Subject: *Re: [Lustre-discuss] Lustre over two TCP interfaces

      thank Michael,
      This is my second step, I will change the lnet with “options lnet
      networks=tcp0(bond0,bond1)” because my machines has 4 nics. I have a
      bond0 and bond1 with LACP. I need to comunicate the clients with two
      network for HA network.
      If the bond0 network is down, the clients can reach the OSS by the
      second network bond1.
      If I change the modprobe with “options lnet
      networks=tcp0(bond0),tcp1(bond1)”, how the clients mount the filesystem
      to reach the OSS by two network?
      *From:* Michael Shuey <mailto:shuey at purdue.edu>
      *Sent:* Tuesday, June 25, 2013 2:14 PM
      *To:* Alfonso Pardo <mailto:alfonso.pardo at ciemat.es>
      *Cc:* lustre-discuss at lists.lustre.org
      <mailto:lustre-discuss at lists.lustre.org> ; WC-Discuss
      <mailto:WC-Discuss.Migration at intel.com>
      *Subject:* Re: [Lustre-discuss] Lustre over two TCP interfaces
      Different interfaces need to be declared with different LNET networks -
      something like "networks=tcp0(eth0),tcp1(eth1)".  Of course, that
      assumes your clients are configured to use a mix of tcp0 and tcp1 for
      connections (with each client only using one of the two).  This is
      really only useful in corner cases, when you're doing something strange;
      if eth0 and eth1 are in the same subnet (as in your example), this is
      almost certainly not productive.
      A better bet might be to use a single LNET, and bond the two interfaces
      together - either as an active/passive pair, or active/active (e.g.,
      LACP).  Then you'd declare networks=tcp0(bond0), give the bond a single
      IP address, and client traffic would be split across the two members in
      the bond more like you probably expect (given the limits of the bond
      protocol you're using).
      Mike Shuey

      On Tue, Jun 25, 2013 at 8:06 AM, Alfonso Pardo <alfonso.pardo at ciemat.es
      <mailto:alfonso.pardo at ciemat.es>> wrote:

          hello friends,
          I need to comunicate my OSS by two ethernet TCP interfaces: eth0 and
          I have configured this feature in my modprobe.d with:
          “options lnet networks=tcp0(eth0,eth1)”
          And I can see two interfaces with:
          lctl --net tcp interface_list
          sa-d4-01.ceta-ciemat.es <http://sa-d4-01.ceta-ciemat.es>:
          ( <>)
          npeer 0 nroute 2
          sa-d4-01.ceta-ciemat.es <http://sa-d4-01.ceta-ciemat.es>:
          ( <>)
          npeer 0 nroute 0
          But, the clients only can communicate with the first interface:
          lctl ping
          12345-0 at lo
          12345- at tcp
          lctl ping
          failed to ping at tcp: Input/output error
          Any suggestions how to “enable” the second interface?
          thank in advance

          /Alfonso Pardo Diaz/
          /*System Administrator / Researcher*/
          /c/ Sola nº 1; 10200 TRUJILLO, SPAIN/
          /Tel: +34 927 65 93 17 <tel:%2B34%20927%2065%2093%2017> Fax: +34 927
          32 32 37/

          CETA-Ciemat logo <http://www.ceta-ciemat.es/>

          ---------------------------- Confidencialidad: Este mensaje y sus
          ficheros adjuntos se dirige exclusivamente a su destinatario y puede
          contener información privilegiada o confidencial. Si no es vd. el
          destinatario indicado, queda notificado de que la utilización,
          divulgación y/o copia sin autorización está prohibida en virtud de
          la legislación vigente. Si ha recibido este mensaje por error, le
          rogamos que nos lo comunique inmediatamente respondiendo al mensaje
          y proceda a su destrucción. Disclaimer: This message and its
          attached files is intended exclusively for its recipients and may
          contain confidential information. If you received this e-mail in
          error you are hereby notified that any dissemination, copy or
          disclosure of this communication is strictly prohibited and may be
          unlawful. In this case, please notify us by a reply and delete this
          email and its contents immediately. ----------------------------

          Lustre-discuss mailing list
          Lustre-discuss at lists.lustre.org <mailto:Lustre-discuss at lists.lustre.org>

    Brian O'Connor
    SGI Consulting
    Email: briano at sgi.com, Mobile +61 417 746 452
    Phone: +61 3 9963 1900, Fax:  +61 3 9963 1902
    691 Burke Road, Camberwell, Victoria, 3124

  Lustre-discuss mailing list
  Lustre-discuss at lists.lustre.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20130627/c047a0ad/attachment.htm>

More information about the lustre-discuss mailing list