Overview

This document presents an analysis of the network traffic usage of DataSplice. The following analysis is based on several assumptions about application usage and should only be thought of as a rough estimate of network resource usage. Because most aspects of an installation of DataSplice can be configured to meet particular needs of the user in question, actual network usage could vary significantly from those observed in this analysis. All tests were performed using DataSplice version 2.3.3 and using unmodified scripts and views for Maximo (SQL Server).

This analysis will attempt to establish baselines of bandwidth utilization and total traffic exchanged over the network between the client and the server. Network analysis was performed using the Ethereal protocol analyzer available at http://www.ethereal.com.

Wireless Session

The following packet captures were obtained using DataSplice on an 11mbit 802.11 wireless connection.

First a new session was established with the server. The test user had access to all 45 Maximo views for DataSplice. The client logged in, displayed a list of 10 workorders, and then viewed the details of a workorder, displaying the job plan, entering a failure report, etc. The entire session lasted about 90 seconds.

Network Traffic:

Total Conversation	28552 bytes
Client -> Server	8659 bytes
Server -> Client	19893 bytes

The following traffic utilization graphs show network throughput during the conversation.

Figure 1 - Throughput Graph (Client -> Server)

Figure 2 - Throughput Graph (Server -> Client)

The first thing to note is the initial spike after the client first connects. Because this was the first time the client had connected to the server it needed to update the view information for all 45 views the user has access to. This accounted for the majority of data transferred during the test.

After the views were synchronized we can see the individual transactions between the client and server separated by periods of no network usage. These are the client and server exchanging record information to view and update data. Since the client is limited to displaying 10 records at a time the utilization stays below 1000 B/s.

The next test was performed by reconnecting the client and continuing work. Again, the client displayed a list of 10 workorders and then viewed and edited the details for several of those workorders. The total session lasted about 2 � minutes. The total network usage was much lower than the previous test because the view information did not need to be downloaded.

Network Traffic:

Total Conversation	8606 bytes
Client -> Server	3071 bytes
Server -> Client	5537 bytes

Figure 3 - Throughput Graph (Client -> Server)

Figure 4 - Throughput Graph (Server -> Client)

Here we can see that aside from an initial small spike the traffic is fairly regular during the course of the connection. The combined bandwidth between the server and client averages to less than 300 B/s when the user is actively using the connection.

CDPD Session

The same tests were also performed on a handheld using a Sierra Wireless CDPD connection.

First a new session was created. This took considerably longer than the wireless connection to download the view data (about 2 minutes).

Network Traffic:

Total Conversation	20431 bytes
Client -> Server	8475 bytes
Server -> Client	11956 bytes

Figure 5 - Throughput Graph (Client -> Server)

Figure 6 - Throughput Graph (Server -> Client)

The break that can be seen at the two minute mark represents when the views finished downloading. All the data afterward was for displaying the list of workorders and modifying a workorder.

Next another connection was made, this time without needing to download any view data. Again, the client downloaded a list of ten workorders and then viewed and modified data for a couple of minutes.

Network Traffic:

Total Conversation	22035 bytes
Client -> Server	3267 bytes
Server -> Client	18768 bytes

Figure 7 - Throughput Graph (Client -> Server)

Figure 8 - Throughput Graph (Server -> Client)

For this connection the traffic utilization is very similar to that of the wireless connection, averaging 200-300 B/s. The scale is off a bit from the same graph in the wireless test because the initial peak after connecting cannot be as high on the slower connection.

Conclusions

The previous tests show that DataSplice is well suited to environments with limited network capacity for regular usage. The initial connection to the server has a fairly high cost when the client needs to download all available views. In low-bandwidth scenarios this should be mitigated in two ways:

• Reduce the number of views each user has access to. The 45 views used in the test represent all DataSplice views for Maximo, including Work Orders, Inventory, Equipment, etc.
• Connect the device initially using a higher bandwidth connection. Because DataSplice will cache views once they are downloaded, the first connection can be performed from a docking station or similar, and then subsequent connections will not need to update any views unless the administrator has made modifications.