[ClusterLabs] Antw: Re: Antw: Re: EL6, cman, rrp, unicast and iptables

Ulrich Windl Ulrich.Windl at rz.uni-regensburg.de
Mon Sep 14 13:33:34 UTC 2015


>>> Noel Kuntze <noel at familie-kuntze.de> schrieb am 14.09.2015 um 13:55 in
Nachricht <55F6B5B5.8020107 at familie-kuntze.de>:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
> 
> Hello Ullrich,
> 
>> Actually I don't understand that claim: If packets are delivered in order
>>(mostly), any TOTEM packet has the same change to arrive than any other 

s/change/chance/ # typing error, maybe lack of coffeine in the morning

> packet.
>> While all other communication protocols can actually deal with Ethernet and

> the
>> Internet, TOTEM is the only protocol that can fail even in a switched LAN.
I
>> haven't benn convinced yet that it's not an implementation issue of TOTEM.
>> Instead of telling people to fiddle with their network configuration, I'd
>> prefer putting more efforts into fixing TOTEM.
> 
> I assume with "change to arrive", you mean delay? Or do you mean the 
> ordering of
> the packets?
> Totem behaves like it does because it needs to detect a failed node, afaik.

What totem does it detect network problems when there are none:

# grep ringid.*FAULTY /var/log/messages |wc -l
1981

> This is something that no other protocol you encounter on the internet/LAN 
> is supposed to do.

Definitely not: 0 interface errors on any interface, not communication
problems.

> All of those protocols are either for error reporting (ICMP) or for 
> transceiving of
> data (udp/tcp). UDP obviously has no congestion algorithm, but TCP does.

Even NFS over UDP is much smarter than TOTEM is.

> 
>> The main problem with priorities is who decides what is most important,
>> especially if a medium is shered by many different software stacks and

s/shered/shared/  #se above

>> applications.
> 
> Obviously some type of prioritzation has to be done, or at least should be 
> done,
> because some things *are* more important than others. The only thing that 
> can
> control congestion centrally in a computer system is the interface that 
> controls
> access to it, so it's either the NIC or the software that controls access to

> it, so the network stack
> of the operating system. The problem is a different one when the LAN is 
> bridged, rather then switched, because then
> the transmission of other hosts affects the transmission of one host.

If you have a central authority that can decide on each eand every priority
you are right. I was talking from practical experience...

> - -- 
> 
> Mit freundlichen Grüßen/Kind Regards,
> Noel Kuntze
> 
> GPG Key ID: 0x63EC6658
> Fingerprint: 23CA BB60 2146 05E7 7278 6592 3839 298F 63EC 6658
> 
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v2
> 
> iQIcBAEBCAAGBQJV9rWvAAoJEDg5KY9j7GZY3LkP/3vkEppL48nwlAGVpFIbVIRj
> HpC6usWTFTaS3s20FOBo+60mtGAi6QnDku05WEkcjKrN8rjb0lll8KKAxCAP5ejO
> xofUpmSZp4vs534gpwYotXf8IU4ZwLsF5WEjdtVc0AoVk99TNwS8g7P2eGRybvxy
> Qdr+C2I99n4iqr93MRjDRRZj5S6t+PICr7s2hRrGrNIiSO0XJJdnoJWYR2g3DlPi
> 9tw2nyb832Pe3eusRqBdXN1lDEw8Amr2apjW6yGlNKlbaVe/TbcxZg4qnuPQtTAa
> Jc9pxItG31ZGG6G3SyzQuU2VG1DUGfyqUBAKv//oQtlb8YEklYHfzvhUvf4/XTJn
> 5Zcv6IVoTUVVexB6bmQ6sHxbsXpHrb7Y+uViqVNEogJ66I4kTi9jo7DxxW3Mjsct
> TSMjGAWEdmhi1KKONuCnqLMvyVdqdF/4VKZhJ6P2NaVQpk/8zXXrp1Q0zJmfupV6
> awQXvwRdAwM4KP+G94KxjFn8J7cuC3a6Hk2LuQp2OL/2IEliN5p8+R0lii6eVev4
> n+wVsgLve/JHMBghNhJTf5Fs6+lUsgOOYt4RK3/gqAFuktE53XqmwdMVjl3yelXR
> UR5J3GxQ5AbuhzetbVn1HIVMfOzwjzgW8vjcWmkmB01tOKXyvpyWRjFP6HawLxCh
> kWHwsh6S+7OxJ0Oijrs5
> =CyyI
> -----END PGP SIGNATURE-----
> 
> 
> 
> _______________________________________________
> Users mailing list: Users at clusterlabs.org 
> http://clusterlabs.org/mailman/listinfo/users 
> 
> Project Home: http://www.clusterlabs.org 
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf 
> Bugs: http://bugs.clusterlabs.org 







More information about the Users mailing list