'HSOR\LQJWKH3UR/LDQW&OXVWHUIRU1HW:DUH
0096-0699-A
Tuning NHAS
Memory Requirement Calculation
Since NHAS runs on NetWare 4.11 and NetWare 4.2, most of the performance tuning for a
cluster is handled by tuning the OS itself. However, the fact that NHAS has two modes of
operation adds a twist to system requirements. Specifically, the optimal amount of RAM that
should be installed in each clustered server will vary drastically between costandby and failover
modes. As with NetWare itself, the recommended amount of RAM depends on a number of
factors: size of the internal drive(s), size of all shared volumes, the applications and name spaces
loaded, number of users, etc. A good reference article on server memory calculation can be found
on the Novell web site at:
http://developer.novell.com/research/appnotes/1995/November/03/index.htm
There is also a DOS based server memory calculator available through the Novell web site called
SMEM.EXE. Whichever method you choose to calculate RAM requirements, remember that you
should calculate using a failover condition, where all shared volumes are mounted on one server,
as the basis.
Caching Concerns
As a matter related to performance, NetWare uses a file cache to increase server performance.
However, when this cache is used and a server fails, information that is contained in the file cache
and not yet written to disk is lost. Configure the NetWare file cache to balance your performance
needs against the possible loss of critical data.
For more information on tuning NHAS to your system requirements, see the documentation
accompanying the software.
Technical FAQs
What happens to a client running server based applications when a server failover occurs?
NetWare High Availability Server can use scripts or the NetWare Application Launcher,
a Z.E.N.works component, to restart any server-based applications that may have been
shut down when the server failed. These applications will restart on the surviving server,
allowing the client to continue working from the other server node. Most client-based
applications will continue to function normally, with a slight pause while the volume is
reconnected through the persistent links. Some applications will require reloading to
continue operation.
What happens when the client is running local applications to work on a document located on the
server node that failed?
Most applications will continue to function normally. Clients with a stateful connection,
such as IPX, will not be able to reconnect the volume that was previously in use since the
volume is referenced differently in the NDS tree. The client will not have to reconnect,
but will have to select the new volume object under the other server node in the NDS
tree.
