HP OpenVMS Systems

ask the wizard
Content starts here

TCP/IP (UCX) cluster alias load balancing?

» close window

The Question is:

 
We have a OpenVMS 7.1 cluster of two VAX 4108's running UCX 4.2.
 
I want to enable cluster load balancing from PCs attaching to the cluster
via TCP/IP using the alias. I have managed to get this working within the
cluster itself and from other VAX's in the same domain. The only way I could
do this was to configure the B
IND resolver on each node and remove the cluster alias record from the host
database, allowing it to be found in the BIND database. I have used
UCX$METRICVIEW and the BIND debug logicals to determine that it is working
as expected.
 
How is it possible to set this up for access from PCs?
 


The Answer is :

 
  What application are you using on the PC systems?  If the application
  is a normal Internet application (SMTP and POP mail, LPD printing,
  TELNET, FTP, etc.), then a PC client receives the same benefit from
  nameserver-based load balancing as would any other client.
 
  On the other hand, if you are using PATHWORKS (Advanced Server), then
  effective cluster load balancing requires somewhat more effort.  The
  Wizard has been told the PATHWORKS group is planning to address this
  in a future release.  Meanwhile, it may be possible to achieve some
  load balancing through careful configuration.)
 
  The key to handling load balancing in PATHWORKS (Advanced Server) is
  how the client resolves the netbios name specified.
 
  As long as all clients are in the same subnet as the PATHWORKS cluster
  (hardly a common scenario), then allowing clients to send UPD broadcasts
  to resolve the PATHWORKS cluster alias name (so each cluster member
  running PATHWORKS sees the name request) will get you load balancing
  (based solely on PATHWORKS' load-balancing scheme) and will also
  provide fairly transparent failover.
 
  When you expand to a WAN environment, the broadcast approach gets rather
  more messy, obviously.
 
  The PATHWORKS cluster alias name is not tied to any IP address.  In
  order for remote systems to use the alias name, that name needs to be
  defined in some (NetBIOS) name table (because the broadcasts won't pass
  the routers), statically mapping the name to a specific address.  Some
  customers are using the UCX cluster impersonator address for this
  purpose - but this pretty much eliminates any load-balancing.
 
  A similar problem arises for WINS clients - how do you define the
  PATHWORKS cluster alias name in the WINS database when there is no
  IP address assigned to it?  Many customers define a static, multi-homed
  entry, including the IP address of each cluster member running PATHWORKS
  in the list.  But this isn't bullet-proof either (but seems to be the
  least painful approach), especially for Microsoft Windows 95 clients.
 
  Finally, don't even try using the BIND "load-balancing" mechanisms and
  setting the Windows clients to resolve NetBIOS names via DNS.  Well,
  perhaps if you have all Windows 95 clients -- this approach is rather
  problematic for Windows NT clients, at best.  In fact, most of it doesn't
  work due to the different manner in which Windows NT clients, when
  resolving a NetBIOS name via a DNS query, attempt to connect to the
  PATHWORKS server.
 

answer written or last revised on ( 22-DEC-1998 )

» close window