[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference lassie::ucx

Title:DEC TCP/IP Services for OpenVMS
Notice:Note 2-SSB Kits, 3-FT Kits, 4-Patch Info, 7-QAR System
Moderator:ucxaxp.ucx.lkg.dec.com::TIBBERT
Created:Thu Nov 17 1994
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:5568
Total number of notes:21492

5288.0. "Another load balancing problem?" by HGOSPS::KENNETH () Fri Feb 28 1997 05:44

Hi,

There is a customer report a problem on UCX load balancing which is quite 
similar to entry number 4889 of this conference.  

The customer has an Alpha 8400 and an Alpha 7750 to for a cluster with the
alias node name cabs2000.

Everytime when he UCX > show host will give the lower rating node as the
first node display.  Please find more information below.
 

UCX Version V4.1 and VMS version is V6.2-1H3.

Is there any problem in the setting that affect the load balancing
result?

Thanks for your help in advance.

Kenneth Leung

p.s.  I already know that the 2 minutes interval, I monitor the case for 
      a very long time.

==============================================================================

UCX>show host CABS2000
  
	BIND database

Server: 192.168.1.85	DALLAS  ----- primiary DNS server which is not member
                                      of the cluster.

Host address		Host name

128.3.1.3		CABS2000.ITB.HCL.COM
128.3.1.4		CABS2000.ITB.HCL.COM


$MC UCX$METRICVIEW

Host					Rating

128.3.1.3	ATHENS			60 --- secondary DNS server
128.3.1.4	VIENNA			89

$sho log ucx$bind*  (on the cluster node)


  "UCX$BIND_DOMAIN" = "ITB.HCL.COM"
  "UCX$BIND_RETRY" = "...."
  "UCX$BIND_SERVER" = "........"
  "UCX$BIND_SERVER000" = "192.168.1.85"
  "UCX$BIND_SERVER001" = "127.0.0.1"
  "UCX$BIND_STATE" = "........"
  "UCX$BIND_TIMEOUT" = "...."
  "UCX$BIND_TRANSPORT" = "UDP"


$sho log ucx$bind*  (on primiary server DALLAS)

  "UCX$BIND_DOMAIN" = "ITB.HCL.COM"
  "UCX$BIND_ENABLE" = ".1.."
  "UCX$BIND_RETRY" = "...."
  "UCX$BIND_SERVER" = "........"
  "UCX$BIND_SERVER000" = "192.168.1.85"
  "UCX$BIND_STATE" = "........"
  "UCX$BIND_TIMEOUT" = "...."
  "UCX$BIND_TRANSPORT" = "UDP"
  
T.RTitleUserPersonal
Name
DateLines
5288.1Few more questionsHGOSPS::KENNETHMon Mar 03 1997 06:2623
Hi,

Regarding to .0 I would like to ask few more questions.
Please forgive my ignorance, this is my first time to deal with UCX load 
balancing.  All my knowledge is from the UCX manuals.

1) If I start the process UCX$METRIC, does it means that the load 
   balancing on the DNS server is turn on?

2) Do we still need to turn off Round Robin but setting
   the logical UCX$BIND_ROUND_ROBIN_OFF to "true"?
   
3) Why does the Telnet login goes to the node with much lower rating, is
   there any configuration issue?  

4) Can anyone tell me the load balancing rating algorithm?

5) Any articles (steps by steps) to set up DNS and load balancing for 
   UCX, besides the manuals.

Thanks for your help in advance.

Kenneth Leung
5288.2some answersUTRTSC::KNOPPERSOswald KnoppersMon Mar 03 1997 06:4428
>1) If I start the process UCX$METRIC, does it means that the load 
>   balancing on the DNS server is turn on?

No, you will also have to configure the bind server using ucx 'set config
bind/cluster=xxx' commands.

>2) Do we still need to turn off Round Robin but setting
>   the logical UCX$BIND_ROUND_ROBIN_OFF to "true"?

If you want load balancing you don't use this logical.

>3) Why does the Telnet login goes to the node with much lower rating, is
>   there any configuration issue?  

Probably because your configurations is not setup as it should.

>4) Can anyone tell me the load balancing rating algorithm?

It is almost the same algoritm as LAT uses for its rating.

>5) Any articles (steps by steps) to set up DNS and load balancing for 
>   UCX, besides the manuals.

Not that I know of.

Regards,

Oswald
5288.3Primary DNS server is not part of the cluster, ok?HGOSPS::KENNETHTue Mar 04 1997 05:2622
Hi,

I find that the primary DNS server is not one of the member of the
TCP/IP cluster.  Will it cause the problem?

UCX>show host CABS2000

        BIND database

Server: 192.168.1.85    DALLAS  ----- primiary DNS server which is not member
                                      of the cluster.

Host address            Host name

128.3.1.3               CABS2000.ITB.HCL.COM
128.3.1.4               CABS2000.ITB.HCL.COM


Thanks again,

Kenneth Leung

5288.4should be no problemUTRTSC::KNOPPERSOswald KnoppersTue Mar 04 1997 05:437
>I find that the primary DNS server is not one of the member of the
>TCP/IP cluster.  Will it cause the problem?

No. Not as long as the bind server is able to communicate with the metric
servers on the cluster nodes.

Oswald
5288.5No metric information found on primary DNS server.HGOSPS::KENNETHWed Mar 05 1997 05:1733
Hi Oswald,

Thanks for your help.  However, I still have some uncertainties.

In ".2" you said:

>1) If I start the process UCX$METRIC, does it means that the load
>   balancing on the DNS server is turn on?

>No, you will also have to configure the bind server using ucx 'set config
>bind/cluster=xxx' commands.
 
But from ".4" you mention there is no need for the primary DNS server 
belongs to the members of the TCP/IP cluster.


My question is :

1) Does it means that we need to type ucx 'set config bind/cluster=xxx' 
   commands on the primary server even it is not one of the cluster members?
	
Moreover, I found that on the customer's site, when we show the metric 
figures using " mc ucx$metricview", there are no information displays even 
the customer has already started the metric server (@UCX$METRIC_STARTUP).

Does it normal?  On our testing system, I found that even I do not start
the metric server, I can still see the other node's rating by typing
" mc ucx$metricview".  Do you know how the DNS server exchange or 
receive metric informations from the cluster members?

Thanks again.

Kenneth Leung
5288.6works for meUTRTSC::KNOPPERSOswald KnoppersWed Mar 05 1997 09:5525
>1) Does it means that we need to type ucx 'set config bind/cluster=xxx' 
>   commands on the primary server even it is not one of the cluster members?

You will want to do this on all bind servers (primary and secondary).

>Moreover, I found that on the customer's site, when we show the metric 
>figures using " mc ucx$metricview", there are no information displays even 
>the customer has already started the metric server (@UCX$METRIC_STARTUP).

No, this is not normal. On our (two node) testcluster I see the following:

$ mc ucx$metricview
Host                                                    Rating
----                                                    ------
128.1.4.12      aussie                                    49
128.1.4.14      tasman                                    50
$

I get the same result if I issue this command on a node outside the
cluster. Did you verify that the metric server process was actually
running?

Regards,

Oswald