DE4204113C1 - Measurement data compression method for communication and control processes on distribution line - defining identification classes for data sets, and protocolling repeated sets - Google Patents

Abstract

The data compression method differentiates communication and control processes only in reference to one component or several components together, where one or more components are compounded into communication units. The control processes which characterise communications processes on the distributing line and which refer exactly to one communications unit are compounded into condition classes. Successive condition classes that repeat are detected only once, and are characterised in a unique way w.r.t. this repetition. A fixed succession of condition classes is recognised as a new class. The distribution line may be a data bus, the communication units addresses, and the condition classes data bus conditions. USE/ADVANTAGE - For computing systems with software, hardware or hybrid monitors. All events relevant to the measuring of control and communications processes on a distributing line can be detected.

Description

As a rule, when developing computer systems Bus lines, such as standardized bus systems, are used. These standard buses connect the various computer compos nents, such as B. memory, CPU and peripherals together.

Some such standard buses are the PCAT bus, the EISA bus, the Multibus I, the Multibus II, the VME bus, the SCSI bus and the S-Bus. See also: "Joseph di Giacomo: Digital Bus Handbook. Mc Graw-Hill, 1990 ".

The increasing performance requirements for computing to be developed Computer systems are usually found in higher processor clock rates or in broader data words, i.e. H. for example in the transition from 16 bit to 32 bit architectures. That has as a result that on the standard buses used, the over have a limited transmission capacity, one always increasing communication volume takes place.

With increasing computing power, these buses become narrow pass in the computer systems. There is therefore a be vested interest in determining to what extent a Bus already by an existing computer implementation is loaded or, if the bus is already overloaded, firmly which of the bus partners is causing this overload.

The monitoring of such bus lines can be done by Hard goods or software bus monitors. Software monitors are based on programs being instrumented. Such Programs can e.g. B. be idle programs who started who  when the CPU is idle, IO operations or write and read access to the system memory.

Instrumentation means that the respective program or the perform the corresponding operation with a little extra baren program code is provided, which when calling the respective current routine or the respective program causes a Message is generated that is sent to a predefined bus address is dispatched. There is a counter connected to the bus, who evaluates and counts up these addresses.

With a software monitor you can determine how often certain programs or routines are executed. A disadvantage is that the number of executions of a particular program not necessarily with the load generated thereby on the Bus correlates. That means a software bus monitor is not suitable for providing information about the bus load. Further Disadvantages are that you have to introduce software monitors Access to the respective source code of the programs and Rou tinen needed and that additional through a software monitor CPU load is generated. It is generally assumed that this load is in the 1% range. However, this is critical when software monitors are used in real-time systems should.

Event comes as further options for bus monitoring triggered hardware monitors. You monitor that Addressing certain predefined bus addresses that are with corresponding communication processes connected to the bus are. For writing to system memory, e.g. B. Addresses 1000 to 2000 have been assigned. Now become one of these Addresses from the bus addressed, so the hardware monitor this firmly and note the process together with the time at which this process took place. The result nis is stored in a buffer memory that is saved with a mas sensor memory is coupled.  

Disadvantages of hardware monitors are:

• -The elaborate process with the timestamp that additional 48 to 96 bits required for the recorded events.
• -The immense memory requirement when recording all instead finding events on the bus.

One can imagine that with data of 96 bits per Event and bus clock rates of 15 MHz quickly at the limits also a large magnetic disk in the gigabyte range is. This makes it necessary to use hardware monitors that event-triggered, considered only certain bus operations can be tet.

Even then it is only because of the memory problem described possible to record them in the millisecond range. There with event-triggered hardware monitors not all events nisse that take place on the bus for a long time space can be observed, are also no statements about utilization or overloading of the bus is possible.

The same applies to sampling hardware monitors that use the bus state in equidistant or random time intervals mes sen. If the time interval is chosen too small, the result is also an immense storage requirement.

In the foregoing, reference is made to: Ferrari, D., Ferazzi, G., Zeigner, A. "Measurement and Tuning of Computer Systems, Prentice-Hall, Englewood-Cliff, NJ 1983 "reference genom men. In "IBM Technical Disclosure Bulletin Vol. 31 No. 1 June 1988 "describes a method by which errors data can be stored in a log in a memory-effective manner can.  

Data reduction is achieved by depending on the The type of error and the time of its occurrence determined Data such as time, day, month, are not saved. In addition, similar errors are caused by a re Fetch counter recorded.

In "IBM Technical Disclosure Bulletin Vol. 34, No. 5, October 1991 "describes a process which the Performance of a calculator analyzed by it "Traces" evaluates. There are various stocks parts of the computing system, such as. B. the virtual memory, the logical memory and the physical memory, separately monitored and statistically evaluated. Other possibilities for bus monitoring are currently not known.

The object underlying the invention is specify a procedure for manifolds with which it is possible, all for the measurement of control and Communication processes relevant to the manifold Capture events.

This object is achieved according to the features of patent claim 1 solved.

It is advantageous in the method described that three Data reduction methods used for data collection to those who are not evaluating any aspect of the measurement limit the solution afterwards.

It is also advantageous that the method is special good for long-term load measurement and recording of measurements results from manifolds.

For one, these are the summary of buses Individual addresses for address areas, the division of all  Bus states in bus state classes and the introduction of a Repeat labeling for the occurrence of the same Zu classes in a row.

The status classes can be minimized and represented with s = log ₂ (ZK) digits (bits). ZK denotes the number of condition classes. So you can z. For example, reduce the 40 bits required to record all bus states on the Multibus II to 32 bits when using 32 status classes. This also reduces the memory requirement for recording all events to 1/8 of the original. If these three methods for data compression are implemented in hardware and used directly at the measuring location, the data rates from the measuring location to the buffer memory and also from the buffer memory to the magnetic disk are reduced to 1/8.

Provided that to record a bus condition class eight-bit information units are used, remain, after that only 5 bits are used for the bus status class, three Bit left for a retry counter.  

Summarizing four such events into one 32-bit Word allows, together with those reduced by compression graced speed requirements, the buffer memory in to implement a 32 bit wide memory that is inexpensive can be realized from D-RAMS.

Other developments of the invention result from the Subclaims.

Using an exemplary embodiment in which the Multibus II a Manifold represented in figures, the invention is explained in more detail.

Fig. 1 shows the summary of states of a bus line to state classes.

Figure 2 illustrates data compression by introducing retry flags.

Fig. 1 shows how individual states of a bus, which result from measurement data and characterize the control processes cha, which relate to communication processes on the bus and which exactly one communication unit, that is, one or more components, are assigned to state classes will. In this example, the states of the bus are bus states of the Multi bus II. The theoretically possible number of bus states that the Multibus II can assume is derived from the number of data lines 32 and the number of control lines 8 . This is a total of 40 lines and the number of possible bus states is 2 ⁴⁰ . This number is reduced by a sensible choice of the condition classes to, for example, 32 different condition classes. In the example, all possible bus states in the Multibus II are identified by the start of a transfer BT and the end of a transfer ET or the rest R. Since the Multibus II works according to message passing, only observation of the message address space is interesting for determining the bus load.

For example, the message address space is divided into five partners divided, which can send messages VA, VB, VC, VD and VR.

The messages can be of a kind Affect partners NA, NC, ND, NR or that they are ge are aimed. In this case, a broadcast B takes place.

It can easily be seen from this that data transmissions are identified only by their beginning and end and by the communication units involved and not by their content. When recording the end of such a transfer, it is still necessary to note whether this has taken place without errors FF or with errors FB. This division makes it easier for the measurement person to later evaluate the measurement data. After adding up all the bus status classes described as an example in FIG. 1, 31 is obtained. These can be recorded in 5-bit representation by simply assigning a number to each bus status class.

Fig. 2 shows an example of a communication process on a bus, here the transfer process on the Multibus II is shown how data compression takes place by introducing a repeat identifier.

The presence of the send address B and the receive address A on the bus causes the bus status class VB / NA to be generated. The data transfer on the bus follows. In the representation selected here, up to 8 data packets can be transmitted. After data transfer has ended, status class ET is generated at the end of the transfer. Together with the ET status class, a repeat identifier of up to 3 bits in length is logged, which indicates how many data packets were transmitted between the start of the transfer process and the end of the transfer process. Here in Fig. 2 there are a total of 6 data packets.

Here it is illustrated that the content of the Data packets are relevant, but only their number for ver turn comes. The data compression introduced has the effect that instead of the 7 operations specified by the bus protocol, the would be necessary for the transfer process shown, only three processes are saved in the bus state class representation Need to become.

It should also be noted that the recording of the Control signals SC0 and SC2 - 4 can be dispensed with as they can be derived implicitly from the condition classes.

Further compression can be achieved if the entire transfer, d. H. the consequence of the condition classes VB / NA, Rest (multiple), ET, together into a single condition class is quantified.

Claims (5)

1. Method for data compression of measurement data, which relate to communication and control processes and which occur on a collecting line to which components are connected,

• a) in which communication and control processes are only differentiated in relation to one component or several components by combining one or more components into communication units,
• b) in which control processes that identify communication processes on the bus and relate to exactly one communication unit are summarized in status classes,
• c) in the case of repeating, successive status classes, they are recorded only once and clearly identified with regard to their repetition,
• d) in which a fixed sequence of status classes is identified by a new status class.

2. The method of claim 1, wherein the manifold Bus is where the communication units address ranges and the status classes correspond to bus status classes. 3. The method of claim 2, wherein the operations on the Bus by type, sequence, causer, recipient, amount, Duration and for the control also marked content are. 4. The method according to claim 1 and 3, in which for recording per condition class with repeat identification at least 1 byte is used. 5. The method according to claim 1 and 4, in which for recording 32 bit wide words are used, which in at least one 32 bit wide memory can be stored.