IBM E16RMLL-I - Tivoli Storage Manager, Implementation Manual, Page: 513 / 828

Chapter 15. Performance considerations
483
Clearly, with so many combinations and possibilities, it is not feasible to discuss
all possible combinations of these parameters within the scope of this book. This
book should be taken as advice to follow when faced with performance
questions, so that you do not limit yourself to considering only the Tivoli Storage
Manager tunable parameters covered here.
Performance tuning and bottleneck discovery are complex tasks requiring deep
knowledge of Tivoli Storage Manager, the underlying operating system, and the
attached hardware components. Unless you really know what you are doing, you
should consider hiring someone more experienced for these tasks.
Remember that a high-performance environment without any bottlenecks does
not exist. But as long as you know your bottlenecks, and have reduced them and
the constraints they cause, you are on the right path.
15.1.1 Network benchmarking
Network benchmarking in Tivoli Storage Manager is the process of measuring a
given network’s capacity at a specific time during the day or evening.
Sometimes during a Tivoli Storage Manager deployment our customers have
commented that the product does not perform as expected, that backup times
are very slow, and they ask what is wrong with the backup product. It should be
understood that all backup products are essentially an integration of bottlenecks.
As a result, you need to be able to isolate and address these weak links as
issues are identified.
A very large percentage of these poorly performing configurations are not related
to Tivoli Storage Manager at all. But, how do you prove this when the network
has some busy moments during the day, but reaches complete saturation during
evening hours? Obviously, measurements taken at 11:00 a.m. are not useful to
determine what is going on at 11:00 p.m.
The answer to this very vague problem is to benchmark the network bandwidth
precisely before (or after, or both) the backup starts. Customers normally ask
“Who will do this, and how are they going to time this precise effort to occur at
11:30 p.m.? Our evening operations staff is small and rather busy during these
hours; they do not have time to be running benchmarks.”
Here is how it works: The first task is to generate a data file that is
“sizable”—based on the network over which you are running. For example, if you
are running over a switched 100 megabit (averaging 10 MBps), you would create
a file sized between 256–512 MB. For a 1 Gigabit (1 Gbps) network, use a file of
1 GB or larger. Sun Solaris nodes have a handy command,
mkfile
, which is
perfectly suited for creating exact sized files; otherwise you can use a