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

Conference smurf::ase

Title:ase
Moderator:SMURF::GROSSO
Created:Thu Jul 29 1993
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:2114
Total number of notes:7347

2052.0. "Better way to laod balance OPS than just DNS." by MEOC02::JANKOWSKI () Thu May 08 1997 09:28

    I understand that a normal configuration of OPS on TruCluster 
    Production Server Software is such that each machine will have
    its own address, which the OPS users can use.
    
    For evample for hosts X, Y and Z in the cluster we may allocate
    addresses AX, AY, AZ. Those addresses do not need to be aliases 
    and will not be associated with any ASE service.
    
    Some very rudimentary round robin load balancing can be provided
    by DNS. One can configure a pseudo host called say ops-cluster 
    and give it in DNS three A records - AX, AY and AZ.
    
    This works well if all hosts are up. If one of them goes down
    (say AX) however, then new users directed by round robin DNS 
    to this host will get stuck with a very long timeout.
    
    There must be a better way.
    
    In note 2017.1 Eric Schott mentioned - lbnamed - load balancing named.
    Does anybody know more about lbnamed?
    Any pointers to its distribution site?
    Anybody tried to use it?
    
    I thought of another approach.
    In this approach the addresses for OPS would be aliases - say
    ALX, ALY and ALZ. Each of those aliases would be configured as 
    a separate ASE user defined service (no disks).
    In case of a failure of machine X the service holding the ALX 
    alias would failover to either Y or Z.
    This of course will cause certain imbalance between machines 
    Y and Z - the one running two services will get twice as
    many requests as the other - but users will not have to wait.
    
    Another approach that I could invent would be to use one of the 
    generic TCP gateways used in firewalls, modify it slightly to
    do load balancing and then pass all the traffic through such gateway.
    To avoid a single point of failure this load balancing service 
    will work as an ASE service of course (:-)) on the same physical
    cluster. In fact we had done similar modification here already
    a few years ago for a customer wishing to have a load balancing
    telnet gateway. This a more complex approach but any desired
    load balancing algorithm can be used.
    
    Any comments, thoughts, ideas, war stories?
    
    Regards,
    
    Chris Jankowski                      
    Digital Australia
    
    
    
    of its own
    
T.RTitleUserPersonal
Name
DateLines
2052.1KITCHE::schottEric R. Schott USG Product ManagementThu May 08 1997 13:516
Hi

 To load balance OPS, you need both a hostname and instance name
pair to connect...I don't think you will do this with a
name daemon...typically this has to be coded into the application...

2052.2SQLnet/OPS load balancing ??JANIX::jmhsendmail: The Vietnam of BerkeleyThu May 08 1997 16:1114
Hi,
	well using DNS round robin or a modified DNS server (like
	lbnamed) might be possible in many environments as it
	will require (most likely) to the current DNS layout/setup...
	You might be interested in looking at note 1358.3 in
	EPS::ORACLE which gives some hints on how to configure
	an OPS installation to perform automatically load balancing
	via SQLnet . We've tried this on one of our lab clusters
	and it seemed to work with some more fine tuning 
	required ...

Hth,
	Jan