Re: Gluster and bonding

Jorick Astrego <jorick@xxxxxxxxxxx> · Wed, 27 Feb 2019 10:20:16 +0100

    On 2/25/19 6:01 PM, Alvin Starr wrote:

    On
      2/25/19 11:48 AM, Boris Zhmurov wrote:

      On 25/02/2019 14:24, Jorick Astrego wrote:

          Hi,

          Have not measured it as we have been running this way for
          years now and haven't experienced any problems with "transport
          endpoint is not connected” with this setup.

        Hello,

        Jorick, how often (during those years) did your NICs break?

      Over the years(30) I have had problems with bad ports on switches.

      With some manufactures  being worse than others.

    Hi,
    Have been doing 25 years of infra and I have seen really
      everything break and PDSS (People Doing Stupid Sh*t)

    The NIC's these days are excellent quality and we never had one
      break in 10 years. We do a lot of testing before we put it into
      production and we have had some other issues that have the same
      effect (switch faillure, someone pulling the wrong cable, LACP mis
      configuration). 

    Actually we went from LACP with stacked switches to balance-alb.
      There were more configuration errors with LACP and we had stacked
      switches getting messed up. We now have separate L2 storage
      switches.
    And the GlusterFS developers think it's the best bonding mode for
      their application, so you don't have to take my word for it ;-)
    https://docs.gluster.org/en/latest/Administrator%20Guide/Network%20Configurations%20Techniques/

      best bonding mode for Gluster client is mode 6 (balance-alb),
        this allows client to transmit writes in parallel on separate
        NICs much of the time. A peak throughput of 750 MB/s on writes
        from a single client was observed with bonding mode 6 on 2
        10-GbE NICs with jumbo frames. That's 1.5 GB/s of network
        traffic.
      another way to balance both transmit and receive traffic is
        bonding mode 4 (802.3ad) but this requires switch configuration
        (trunking commands)
      still another way to load balance is bonding mode 2
        (balance-xor) with option "xmit_hash_policy=layer3+4". The
        bonding modes 6 and 2 will not improve single-connection
        throughput, but improve aggregate throughput across all
        connections.

    Regards,
    Jorick Astrego

Met vriendelijke groet, With kind regards,

Jorick Astrego

Netbulae Virtualization Experts 
Tel:  053 20 30 270     info@xxxxxxxxxxx     Staalsteden 4-3A     KvK 08198180
    Fax: 053 20 30 271     www.netbulae.eu     7547 TA Enschede     BTW NL821234584B01

_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users