Using “What-if” modeling
Prerequisite: Observer Expert or Observer Suite
Observer What-if Analysis tools lets you virtually test different network and server configuration scenarios based on actual usage patterns captured from your network. It helps you answer the following types of questions:
Will buying more bandwidth improve response times for my users, or would my money be better spent on faster servers, hard disks, or clients?
How would telecommuting over a modem or DSL line affect a particular user's response time?
How would adding more users to a database application affect server response times?
To display the What-if Analysis for any conversation, right-click any conversation that shows a delayed response time and select What-If Analysis from the popup menu. You can change various parameters in the dialog below the what-if graph and see the results immediately.
The top of the display shows transfer rate, response time, or utilization rate given various network bandwidths, extrapolated from the data measured on your network's bandwidth (which is identified by a reference line. As with any such extrapolation, the more data sampled, the more accurate the predictions.
Tell me more about the What-if Analysis tool
What-If live modeling and analysis offers both a predictive tool for modeling potential response times, utilizations, or packets per second at different network speeds, and also permits you to change different conversational and network metrics to predict changes in performance with the new values.
The What-If Analysis starts with a conversation collected from your network and bases all predictions on your actual network data. Different system formulas are used for different types of systems to be modeled.
To begin your What-If live modeling session, right-click a conversation from either the TCP or UDP Events display and select What-If Analysis.
You can only do What-If modeling on conversations that have a recorded server (the second address in any conversation) delay.
The top of the display will show which stations are currently being modeled. The client is on the left, the server is on the right.
The X-axis of the graph will always display different network speeds. If the data collected was from Observer, a vertical reference line will be displayed showing the network speed at which the data was collected.
The Y-axis will display different values depending on the graph type selected.
A key display will show the different items on the graph and their associated colors.
The items below the graph initially represent the actual data from the captured conversation. Items can be changed to model changes in the network.
Observed Connection Parameters (derived directly from the conversation data collected):
Average Packet Size (Bytes)—displays the average size of the packets sent from the client and the server. Changing these values in the Client or Server options will model changes in network performance.
Latency (mSec)—displays the average latency time as observed in the transaction conversation. Values are shown for packets sent from the client and the server. Changing these values in the Client or Server options will model changes in network performance.
Transaction Packet Ratio—displays the transaction packet ratio of the packets sent from the client and the server.
Utilization from other sources (%)—sets the network utilization to simulate. This would be beyond the current conversational conditions recorded, and only changes the modeled values if the option to Include utilization from other sources in What-If Analysis is checked in the Expert Global Settings, What-If tab setup.
User-Defined Parameters are initially set in the Expert Global Settings, What-If tab. Changing the values here will only affect the current calculation and will not be preserved for subsequent modeling sessions.
Graph type—changes what modeling results will be displayed in the graphic view. Options include Packets/sec, Response time (sec), and Utilization (%). While all three views are related, select the view that displays the option you are interested in.
Simultaneous users—sets the number of users to simulate.
Processing Time (ms)—the amount of time, in milliseconds, that the server or client will take to process the request.
Server Characteristics:
Server type—options include Database, FTP, Level, and web servers. Each different server selection causes the expert to use a different formula suited for the selection. A level server offers a formula for a typical server.
Start thread time (ms)—taken into account when the Server type item is defined as web. The value is the amount of time it takes to process a thread on the server.
Arrival rate (trans/sec)—taken into account when the Server type item is defined as Database. The number of transactions per second that are being requested of the (Database) server.
Maximum adapter card throughput (Mbps)—taken into account when the Server type item is defined as FTP. This item defines the server’s maximum throughput. This may be the rated speed of the adapter, but most likely it is some fraction of the maximum theoretical speed (utilization) of the network. The default of this item is set in the Expert Global Settings, under the What-If tab.
One way to get a value for this option is to run Observer on the server using the packet generation mode. Set the generation rate very high and view the utilization that the server can create using Observer’s utilization modes. The maximum utilization will reflect the NIC card’s ability to generate traffic.
Restore Original Values—resets all values to the initial settings for the analyzed pair.
Set Reference—sets the current graph lines to the reference line. For example, if you change the number of simultaneous users from 1 to 100, a What-If prediction line will be displayed and the original reference line will be displayed. If the Set Reference is pressed, the new What-If prediction line will become the reference line for further What-If modeling.