NX 3.5.0 requires at least 20Kbps (kilo-bits per-second) to let users run a remote Linux session comfortably. This is more or less the bandwidth provided by a 28.8 Kbps modem. NX can work on slower links, like 9.6 Kbps GSM connections, but in such network conditions users can experience a noticeable lag.

Bandwidth is not the only parameter to consider when evaluating the performance of a remote access solution. The latency, determined by the characteristics of the network and by the number of routers traversed by the connection, can also affect the responsiveness of the session.

To verify the exact performance that you can expect from NX and determine how many clients can be reasonably run over the network connection you are planning for the deployment, you should proceed experimentally, by running a sample NX session, verifying the real bandwidth usage by means of the NX Session Administrator  interface's session statistics. You should also ensure that the exact applications that you want to run are tested. As NX compresses the X-Window protocol traffic, different applications (or even different versions of the same application) can show slight differences in performance.

To get the session statistics, open the Session Administrator tool, select the running session that you want to monitor, then click Session -> View total statistics.

The numbers near the end of the 'NX Server Side Protocol Statistics' listing, just after the line reporting the protocol compression ratio, will tell you the bandwidth that the session is using, for example:

548 B/s average, 18 B/s 5s, 140 B/s 30s, 3693 B/s maximum.

The numbers are in bytes per-second. In this example, consisting in a KDE session run over a 512Kbps link, the peek bandwidth during the observed time-frame was 3,693 Bytes per-second (less than 40Kbps), with the average bandwidth being 548 Bytes. You can expect 20 sessions to run over such a 512 Kbps link very comfortably (50,000 Bps / 2,693 Bps), with up to 100 concurrent sessions possible (50,000 Bps / 548 Bps), assuming the same average utilization of the network observed during the test.