[AusNOG] VM throughput to network

Dmitri Kalintsev dek735 at gmail.com
Fri Nov 19 13:48:19 EST 2010


...EoMLPPPoE with byte-wise load balancing...

*hides* ;)))

On Thu, Nov 18, 2010 at 4:51 PM, Sean K. Finn <sean.finn at ozservers.com.au>wrote:

>  Fellow Noggers,
>
>
>
> In My experience,
>
>
>
> Bonding doesn’t make multiple 1gb cables into one bigger virtual cable, but
> it does push different traffic from different sources across different
> links. (In most configurations)
>
>
>
>
>
> **OVERALL**, you can breach 1GB/s for **multiple transfers** to/from **multiple
> hosts** on multiple 1Gbs NIC’s.
>
>
>
>
>
> For single sessions however, it’s extremely difficult to implement. This is
> a fact that is not referred to in much, if any documentation at all. You
> kind of have to read between the lines to figure it out.
>
>
>
>
>
>
>
> I haven’t been able to replicate true bi-symmetric Load-Aggreagation in a
> test lab, with almost unlimited configurations of Kit, switches, hardware
> and different vendors. (Not through lack of trying)
>
>
>
>
>
>
>
> Data To/From  one  single host on one side of the bonding to another single
> host on the other side of the boding wont breach 1gbps, but multiple to
> multiple, the aggregate will, with SOME types of bonding.
>
>
>
>
>
>
>
> I hate linking to Wikipedia but this is a pretty good read:
>
>
>
> http://en.wikipedia.org/wiki/IEEE_802.3
>
> Specifically:
>
> http://en.wikipedia.org/wiki/802.3ad dealing with link aggregation and
> talking about one of the different types. (802.3ad specifically)
>
>
>
>
>
>
>
> Have a read here for the different types of host load balancing
>
>
>
> http://www.linuxhorizon.ro/bonding.html
>
>
>
>
>
> Even though the above link is  a linux article, Teaming INTEL Nic’s under
> windows server 2008R2 with the Intel drivers offers the same policy choices.
>
>
>
>
>
> The most effective poor-mans bonding that I can see is Balance-ALB  ( Which
> is Transmit Load Balancing, and Receive Load Balancing)
>
>
>
>
>
> I **HAVE** been able to replicate EITHER Receve Load balancing, or
> Transmit load balancing from 1 host to 1 other host on a single session
> breaching 1gbps, however never both at the same time.
>
>
>
>
>
> Some configurations support Inbound-only Load balancing from the switch to
> your device, and other the polar opposite.
>
>
>
> That is, 4 Nic’s will bond in one direction, but back the other way, only 1
> NIC will bond.
>
>
>
>
>
>
>
> The hardest part to read in between the fine-print is that it’s more
> systematic Load-Sharing of multiple sessions rather than Load-Sharing of
> single sessions.
>
>
>
> If anyone has in practise managed to accomplish this and to genuinely
> increase the throughput of a single IPv4 connection above the throughput of
> that of a teamed single NIC, and can share with me how, I’ll owe you a beer.
>
>
>
> S.
>
>
>
>
>
>
>  ------------------------------
>
> *From:* ausnog-bounces at lists.ausnog.net [mailto:
> ausnog-bounces at lists.ausnog.net] *On Behalf Of *Bruce Renner
> *Sent:* Thursday, 18 November 2010 1:23 PM
> *To:* Jay Mitchell
> *Cc:* ausnog
> *Subject:* Re: [AusNOG] VM throughput to network
>
> A quick google search revealed this.
>
>
>
>
> http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1004048
>
>
>
> The short answer is yes, ESX can use 2 or more nics to improve throughput.
>
>
>
> Cheers All, Bruce
>
>
>
>
>
> On Thu, Nov 18, 2010 at 12:44 PM, Jay Mitchell <jay at miscreant.org> wrote:
>
> It’s configurable at the VMWare side also:
>
>
>
> -        Route based on the originating virtual port ID
>
> -        Route based in IP hash
>
> -        Route based on source MAC hash
>
> -        Use explicit failover order
>
>
>
> --Jay
>
>
>
> *From:* ausnog-bounces at lists.ausnog.net [mailto:
> ausnog-bounces at lists.ausnog.net] *On Behalf Of *Daniel Thoroughgood
> *Sent:* Thursday, 18 November 2010 11:12 AM
> *To:* ausnog
> *Subject:* Re: [AusNOG] VM throughput to network
>
>
>
> I believe ESX  load-balances outbound traffic based upon destination MAC
> address.  If traffic is mostly going to a default gateway, or a storage
> device, then it's unlikely to be moved over more than one link.
>
>
>
> Incoming traffic depends on the switch.
>
>
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net
> http://lists.ausnog.net/mailman/listinfo/ausnog
>
>
>
>
> --
> *Bruce Renner
> *Hosted Services
>
>
>
>  *Phone:* 07 3809 2999 |* Fax:* 07 3809 3999
> *Address:* Unit 2 / 55 Tradelink Rd, Hillcrest, 4118
> www.bettacomputers.com.au
>
>
>
>
>
> Note: This message may contain privileged and confidential information that
> is the property of the intended recipient. The information herein is
> intended only for use of the addressee. If you are not the intended
> recipient, then you are requested to return e-mail to Betta Computer
> Services Pty Ltd and destroy any copies made. Copying or disseminating any
> of this message is prohibited. Any views expressed in this message are those
> of the individual sender and may not necessarily reflect the views of Betta
> Computer Services Pty Ltd.
>
>
>
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net
> http://lists.ausnog.net/mailman/listinfo/ausnog
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20101119/54c1044b/attachment.html>


More information about the AusNOG mailing list