| << Previous Tip | > Tips Table < | Next Tip >> |
Downing the Server: |
|
The fact that NetWare attempts to dynamically tune itself to ever
changing loads is a much promoted feature. While this capability
can be beneficial in certain scenarios, most real world environments
will not receive the benefits such a feature is intended to provide.
For that reason, and others, it can prove advantageous to
periodically down and then restart the Server. Merely downing and
restarting the Server without any prior planning or tuning can
adversely impact the responsiveness of the Server during an initial
period while it tunes itself to the current load. Therefore, the
subject of when or if to down and restart a Server can evoke
arguments for both viewpoints, both being fully credible.
So what is the best advice?
The answer is best determined by the network administrator, but
only after careful analysis. Often, the answer is a combination
of both perspectives...
Although a properly tuned Server rarely needs to be downed and
restarted, almost all Servers should be downed and restarted
periodically to reclaim fragmented memory. While Novell has
dramatically improved memory management in NetWare v4, there
are still cases where memory gets fragmented in the Server.
Some of the problem lies in Novell's reallocation of memory
for internal tables (such as conection tables, FAT tracking,
NDS tables, directory buffers, packet receive buffers, routing
information, etc.).
Another problem is that some NLMs are not designed to properly
manage their memory requirements. Many such NLMs will dynamically
allocate and free a block of memory repeatedly, often in very
small sizes, for temporary storage rather than allocating and
designating a fixed memory block for use, as required. The
repeated allocation and release of such small blocks of memory
can create significant fragmentation when combined with other
memory allocation and reallocation requirements.
Memory fragmentation aside, proper tuning of a Server is another
challenge network administrators must face. Since NetWare will
dynamically tune itself to the ever changing loads it encounters,
it will most often become tuned to worst case scenarios or
exception events. Such tuning does not necessarily guarantee
best overall performance.
Is the load placed on the Server by doing a full Backup or an
Anti-Virus scan representative of normal operations? Chances
are that it is not. Yet, both applications are widely used in
real world environments and impact the Server in such a way as
to cause it to tune itself to the load.
So how can a network administrator maximize the Server's dynamic
tuning capabilities and protect against such adverse tuning?
The key is in the SET Parameters provided in NetWare. The SET
Parameters are most often used when a network administrator
encounters a bottleneck or notices that a previously defined
maximum value has been reached. In such cases, the network
administrator, on their own or at the advice of others who may
not be fully versed on the specifics of the environment, will
bump the values without understanding the full impact of the
adjustment. These types of arbitrary changes can also prove
to generate a negative impact on the Server.
Most of the SET Parameters which are specific to the tuning
of the Server have minimum and maximum values. When properly
defined, the minimums insure that the Server restarts in a
pre-tuned state while the maximums define the limits to Server
self-adjustment to protect against adverse tuning. The challenge
is to validate the optimal minimum and maximum levels then specify
them in the STARTUP.NCF or AUTOEXEC.NCF file.
The following describes one method which can be used for optimal Server tuning:
2. Prior to initiating any Backup, Anti-Virus scans, or
other such periodic events which create non-typical
Server load, record how the Server has tuned itself.
Record the current, tuned values for the Directory
Cache Buffers, Packet Receive Buffers, and
Service Processes.
- The current, tuned values for the Buffer statistics
should become the new minimum values for their
appropriate SET Parameter.
Note: Use the MONITOR NLM or some other utility, such as avanti's NConsole Server Trends Analysis utility, to view the current levels. NConsole can generate SET Parameter reports and perform SET Parameter deviation or variance reports between Servers to make this task easier.
3. Add the appropriate SET Parameters with the new
minimum and, if appropriate, maximum values to the
STARTUP.NCF or AUTOEXEC.NCF file, as appropriate.
4. Repeat steps 1 through 3 for a few days until the
values appear to have stabilized.
5. Calculate the maximum values to represent 125% of
the minimum values (rounded up), then add the
associated SET Parameters to the STARTUP.NCF or
AUTOEXEC.NCF file, as appropriate.
6. Perform these tuning procedures at least once per quarter or whenever any significant load or configuration changes occur. Be sure to start with the already determined minimums and increase the maximums to allow for increased load tuning.
Serious tuning requires long-term trends analysis and capacity
planning that can help identify bottlenecks before they occur,
while insuring optimal performance and configuration. Since
Novell does not provide such tools within NetWare, add-on
utilities must be considered for those wishing to proactively
monitor and plan for their network.
|
| << Previous Tip | > Tips Table < | Next Tip >> |