From: Dr. Martin Körfer (koerfer@mpch-mainz.mpg.de)
Date: Fri Jul 01 2005 - 04:46:39 EDT
The only answer of Tom Smith confirmed my proceeding.
I only used to little differences in the weights.
Thanks Tom
Martin
-----------------------------------------------
Hello Martin.
You seem to be using the right approach. Just be sure that you
raise both rpri and selp. rpri affects which member receives
incoming packets. selp affects which member(s) service the
requests received, regardless of where they arrive.
For single-instance services, I use the following in the
start section of a CAA script:
cluamgr -a alias=[alias],rpri=100,selp=100
cluamgr -r start
and in the stop section:
cluamgr -a alias=[alias],rpri=1,selp=1
cluamgr -r start
and in /etc/clu_alias.config on all members:
/usr/sbin/cluamgr -a selw=3,selp=1,rpri=1,join,alias=[alias]
If you have RIP listeners on the network, you have to increase
rpri by enough to cause the routing metric in /etc/gated.conf.member[n]
to be lowered. A small delta in rpri will affect the arp master
for the alias, but will not affect the routing metric. The possible
metric values range from 14 (rpri=1) to 10 (rpri=100), so there
are steps alomg the way.
-Tom
Dr. Martin Körfer wrote:
> Hi managers,
>
> we run a 2-node TruCluster V5.1a, PK6:
>
> node 1: 1 EV6.7-Processor, 1280 MB Memory
> node 2: 4 EV6.7-Processor, 8 GB Memory
>
> running "sendmail" through the "Internet Express"-Software-package (provided
for
> alpha-clusters) as a "multi-instance service" on both nodes, I want
> to run it preferably on the "4-processor-node2" and only performing failover
> to -node1-, if -node2- shuts down.
>
> I tried to succeed by running the service through creating the cluster alias
> "mailer", to whom I provided high priority on -node2- and low priority on
> -node1-.
>
> However, if -node2- is used by other applications (NFS-Service,
compute-service)
> only for a short time, "mailer" switchs to -node1- and also if showing bad
> performance it does not switch back to -node2-.
> Have I used to litlle priority-values? Which values would you advice to
receive
> the desired behaviour ??
> Is their any possibility to restrict switching only to the case of "shut down"
> of -node2-??
> Is there any other possibility to restrict multi-instance services (under
normal
> conditions, except ailover) to a special node (perhaps through the
"Distributed
> Lock Manager")??
>
> I know: "read the fucking manaual"
> But the Informations therein are so overwhelming that I would like to here
about
> a guidance of experienced mangers on this field.
>
> Any help would be appreciated
>
> Thanks in advance
>
> Martin Körfer
>
>
> --
> Dr.Martin Körfer
> Max-Planck-Institut für Chemie
> Elektronik
> J.J.Becherweg 27
> 55128 Mainz
> Tel.: -49-6131-305488
> Fax: -49-6131-305318
>
> -------------------------------------------------
> This mail sent through IMP: diotima.mpch-mainz.mpg.de
>
> ----------------------------------------------------------------
> This message was sent using IMP, the Internet Messaging Program.
>
----- Ende der weitergeleiteten Nachricht -----
-- Dr. Martin Körfer MPI f. Chemie EDV J.J.Becherweg 27 D-55128 Mainz Tel. -49-6131/305541 Fax -49-6131/305318 ---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program.
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:50:20 EDT