Deep Trace Tool

"REALTIME" Performance Trace Tool based on off the shelf components integrated with custom designed hardware and software.

• Best choice for collection of data used in Trace Driven Simulations.

1. Eliminates the problem of inadequate logic analyzer depth, which results in traces of insufficient length.
2. Eliminates the need to stop the system while waiting for the analyzer to unload data.
3. Eliminates the need to stop the clients driving the System Under Trace (SUT).
4. Eliminates the need for any software modifications or "hooks", which can distort results.
5. Eliminates the dependency on a particular operating system or software package.

The scalable "Deep Trace Tool" is multi-processor capable and consists of eight basic hardware components:

1. 
   Trace tool data probe

2. Pre-Processing Unit
   Handles level conversion of signals, data pipelining, event counting, etc.

   
   

3. Realtime Display unit
   Utilizes output of numerous counters in the Pre-processor Unit to display counts, bar charts, and other graphic elements relating to bus utilization in real time, on a stand alone PC (screen updates approximately every 500 milliseconds). Sample charts:
   

4. Realtime Data Storage Unit, consisting of a high speed RAM storage system.
   

5. Validation Unit, made up of a second data probe, accompanied by a logic analyzer of medium depth.

6. Intermediate Storage Unit, made up of two to four small servers optimized for fast data transfer.

7. Long-term Data Storage Unit, made up of two to four servers optimized for data storage.

8. Post Processing/Modeling Unit, made up of one to two servers optimized for processing and simulations

While the focus is typically on the processor or system bus, the tool can be made flexible by identifying other points of interest within the target system and procuring or designing probes to enable data collection.

Some typical collection sites of interest are:

• The CPU bus
• The PCI bus
• Cluster Controller

Some typical examples of trace content from the system bus:

• Address trace
• Instruction trace
• Branch Taken trace
• Event, or "bus cycle" trace

During a typical usage scenario, the probes are installed in the SUT (perhaps a heavily loaded server optimized for a particular benchmark) and connected to the Pre-Processing Unit.

The Pre-Processing Unit filters and packages the data, sending it downstream to the Realtime Data Storage Unit, while simultaneously processing the data stream and counting events of interest. The counter data is split off from the path to the storage devices and sent to the Realtime Display Unit for instant viewing.

The bulk of the data is then passed to the Realtime Data Storage Unit via a high speed data connection and as the data arrives, an offload begins to the Intermediate Storage Unit. As soon as all the data has been downloaded from the Realtime Data Storage Unit, the tool is available for more data collection. The Intermediate Storage Unit typically holds data for ten to twenty traces and downloads data to the Long-term Data Storage Unit on an overnight basis.

The Long-term Data Storage Unit must contain sufficient storage to hold and allow data manipulation of all the traces in current use.

The Validation Unit works in parallel with the Pre-Processing and Realtime Storage Units to collect an independent subset of samples which can be used to validate the data collected.

For situations requiring accurate performance data, but not requiring deep traces, it is possible to implement only the Trace Tool Data Probe, the Pre-Processing Unit, and the Realtime Display unit. These portions of the tool can yield instantly available, software selectable, statistics and results relating to system activity. Such results can be helpful in performance tuning, system design, and cluster integration.

Comments to: