Skip to contentSkip to conversation

TEQL: 2 Mbit eth1 + 2Mbit eth2 = 1Mbit teql0]

[email protected]: SV: Daniel Bergqvist [email protected] 1 participant 1 message Oct 11, 2000

lore
Share & tools
lore

Daniel Bergqvist [email protected]

author
#0 lore raw

<PRE>Now I understand why it doesn't work.

I will look into it and try to find a solution.

Thanks for all help
Daniel

&gt;<i> -----Ursprungligt meddelande-----
</I>&gt;<i> Från: <A HREF="mailto:[email protected]">[email protected]</A> </I>&gt;<i> [mailto:<A HREF="mailto:[email protected]">[email protected]</A>]För bert hubert </I>&gt;<i>
</I>&gt;<i> On second thought, there is no packet loss. This is expected behaviour, it
</I>&gt;<i> appears, see <A HREF="http://www.kernelnotes.de/kt/latest.html:">http://www.kernelnotes.de/kt/latest.html:</A> </I>&gt;<i>
</I>&gt;<i> Alexey Kuznetsov was critical of this explanation, and said that multipath
</I>&gt;<i> routing worked &quot;perfectly when you need to split load on servers
</I>&gt;<i> talking to
</I>&gt;<i> enough large number of clients. Any
</I>&gt;<i> http server is good example.&quot; He added that Andi's suggestion of the
</I>&gt;<i> existing eql, teql and bonding devices, would introcude &quot;even
</I>&gt;<i> worse problem
</I>&gt;<i> of strong tcp reordering. Actually,
</I>&gt;<i> experiments show that load balancing works only in the
</I>&gt;<i> situations, when
</I>&gt;<i> congestion window is bounded by 3 packets. If it is not made artificially,
</I>&gt;<i> it occurs automatically on each connection
</I>&gt;<i> after some amount of excessive retransmissions. Total single TCP
</I>&gt;<i> connection throughput is never better in this case. Actually, it hints to
</I>&gt;<i> the thought that &quot;true load blalancing&quot; has to
</I>&gt;<i> involve tracking connections and avoiding reordering TCP packets.&quot;
</I>&gt;<i> There was no reply to this, but there was a bit of implementation
</I>&gt;<i> discussion
</I>&gt;<i> elsewhere, along the lines of Andi's
</I>&gt;<i> explanations.
</I>&gt;<i>
</I>&gt;<i> ----
</I>&gt;<i>
</I>&gt;<i> You might consider using google a bit to find out about packet
</I>&gt;<i> reordering -
</I>&gt;<i> packets arrive out of sequence on eth1 and eth2, which the kernel
</I>&gt;<i> interprets
</I>&gt;<i> as packetloss.
</I>&gt;<i>
</I>&gt;<i> Regards,
</I>&gt;<i>
</I>&gt;<i> bert hubert
</I>&gt;<i>
</I>&gt;<i> --
</I>&gt;<i> PowerDNS Versatile DNS Services
</I>&gt;<i> Trilab The Technology People
</I>&gt;<i> 'SYN! .. SYN|ACK! .. ACK!' - the mating call of the internet
</I>&gt;<i>
</I>&gt;<i> _______________________________________________
</I>&gt;<i> LARTC mailing list / <A HREF="mailto:[email protected]">[email protected]</A> </I>&gt;<i> <A HREF="http://mailman.ds9a.nl/mailman/listinfo/lartc">http://mailman.ds9a.nl/mailman/listinfo/lartc</A> HOWTO:
</I><A HREF="http://ds9a.nl/2.4Routing/">http://ds9a.nl/2.4Routing/</A>

</PRE>