T 0084/18 (Testing a distributed processing structure/TELECOM ITALIA) 07-02-2022

1. In the absence of any indication to the contrary, the appellant's statement that it would not be attending the oral proceedings is to be understood as a withdrawal of its request for oral proceedings (cf. T 3/90, Reasons 1, and the further decisions cited in the Case Law of the Boards of Appeal, 9th edition, 2019, III.C.4.3.2). The decision can therefore be made without holding oral proceedings.

2. The application

2.1 The application relates to techniques for automatically testing distributed component-based systems, for example grid computing systems, including a "high" number of components (such as a number of servers in excess of ten) and capable of self-adapting the workload distribution as a function of the quality of service and/or business requirements (description as originally filed, page 1, lines 8 to 13; page 4, lines 8 to 13; page 5, lines 29 to 31).

2.2 The invention aims to automatically test the performance capability of an application based on distributed components by involving a set of test procedures adapted to put under "stress" the application in terms of activities performed in a time unit. The tests have the purpose of singling out functional "bugs" or more simply the operating limits of the hardware and software resources that support the application (page 4, line 27, to page 5, line 3).

2.3 The grid arrangement includes a "high" number of servers, each server being associated with a respective "worker" agent. The worker agents generate information signals that are representative of the operating status of the associated server. Each agent is configured as an "actuator" able to perform certain actions on the associated server (such as selectively turning the associated server on and off) or generate a "fictitious" (test) load in order to saturate a specific hardware resource such as a RAM or a CPU (Figure 1; page 9, lines 12 to 19 and 29 to 34; page 7, line 33 to page 8, line 23; page 13, lines 18 to 25; page 14, lines 1 to 20).

A. Main request

3. Lack of novelty over D8

3.1 D8 discloses a real-time monitoring system that is used to detect (significant) performance shortfalls in a distributed computer architecture (high-performance clusters or HPC) running a computationally intensive application of which the computational needs and resource requirements vary significantly at run time. Some of those applications (for instance for automatic target recognition (ATR)) generate a computational/processing load on the HPC computers that varies drastically depending on the processed data. The resources are reallocated among the application components in an attempt to improve performance. It focuses on a class of applications structured as multiple pipelines of data-parallel stages. The authors emphasise that they considered "practical implementations" rather than "simulation and analysis". Their goal was to develop ways to deliver the aggregate computing power of a parallel machine to such dynamic applications as effectively as possible, since the relative mapping of an application's components to a target machine can have an enormous effect on its performance (abstract; section 1 "Introduction"; section 3.1 "Target applications", first paragraph; section 3.2 "Target machines").

3.2 Claim 1 of the application under appeal does not focus specifically on parallel applications but provides a method of automatically testing the performance of applications run on a distributed processing structure including a grid of processing units. However, the degree of parallelism of a distributed application, namely the capacity for the components of the application to parallelise execution of the activities to be performed, is one of the aspects of performance to be tested by the method of the application (description, page 15, lines 17 to 22 and 29 to 32), even though the kind of "performance" is not furthermore specified in claim 1.

3.3 In D8, the performance of the application is monitored, and a detection of a significant deviation from a desired performance or of significant drops in the performance triggers the computation of a new allocation of resources (processors, memory and communication network) based on "application execution profiles" and the most recent "performance history" of the application. The new allocation is then carried out on the HPC platform. D8 describes, among others, the algorithms used for detection, resource allocation and carrying out (abstract; section 1 "Introduction" in conjunction with Figure 1).

3.4 Thus D8 discloses a method of automatically testing performance of applications run on a distributed processing structure including a grid of processing units, as defined in claim 1.

3.5 The examining division argued that D8 disclosed monitoring the performance of two different applications that ran on a grid and adapting their resource allocation when their performance dropped below some threshold (decision, point 2.1) and that therefore the performance of "applications" was tested.

The board will follow a slightly different line of reasoning. The expression "method of [...] testing performance of applications run on a distributed processing structure" does not necessarily mean that the applications are run at the same time on the distributed processing structure and thus tested in parallel, but encompasses methods which test each application independently. This is not different from the method of D8. It was correct, but there was no need to point out, that the adaptive resource allocation (ARA) of D8 was being tested using a synthetic as well as a real application (D8, section 6, "Preliminary results").

3.6 D8 discloses an adaptive resource allocation (ARA) method which is based on a four-step operational model for dynamic resource allocation, comprising monitoring an application performance using a real-time instrumentation system, detecting a significant deviation in performance from desired performance levels, computing a new resource allocation that is likely to improve performance significantly, and effecting the new resource allocation (section 4 "Adaptive resource allocation (ARA)", Figure 3).

3.6.1 The board acknowledges that, in the application, the load to be generated is represented by the number of workflows that the distributed components have to execute and the input parameters associated with the single workflows (page 5, lines 31 to 35; page 13, lines 6 to 15).

However, this does not appear to differ from the "resource allocation" of D8, which concerns the allocation of processors to the stages of each pipeline of an application (section 3.1 "Target applications", second paragraph; section 3.3 "Resource allocation model", first paragraph).

3.6.2 Thus, the method of D8 includes the steps of "running at least one application on said distributed processing structure" (which the board will itemise as feature (a) of claim 1) and "loading said application with processing workload to thereby produce processing workload on said distributed processing structure" (which the board will itemise as feature (b) of claim 1).

3.7 In D8, "allocation models", which capture task execution profiles and allocation constraints, are used to monitor the performance (or different aspects of performance) of an application (section 4.1. "Adaptation Models").

3.7.1 In D8, as every sub-task of a pipeline stage in the application completes its work on a frame, it sends data about its performance on that frame to a real-time instrumentation system (called the Honeywell Scalable Parallel Instrumentation system or "SPI"). This data consists of the frame identifier, task and sub-task identifiers (specific processors being associated with specific sub-tasks, see section 3.3 "Resource allocation model", first paragraph), and the total time spent by the sub-task in computation and intra-task communication (section 4.2. "Real-time monitoring and detection"; section 4.2.1 "Monitoring", first two paragraphs).

The appellant has argued that "[t]he data indicated by the BoA [...] do not relate to the operating status of the processing units", referring to the definition of "operating status" at page 9, lines 20 to 28 of the application as originally filed (letter of reply dated 16 April 2021, point 2.2).

The board notes that this definition states that "operating status" includes "any set of parameters that properly identify [...] the current conditions of operation of a given apparatus", such as the CPU utilisation, the memory utilisation, the number of processes running on the apparatus, the number of I/O operations performed, "and so on". The total time spent by the sub-task in computation and intra-task communication is considered by the board as "a current condition of an apparatus".

3.7.2 The method of D8 thus includes the steps of sensing the operating status of said processing units in said distributed processing structure under said processing workload and producing information signals indicative of said operating status (which the board will itemise as feature (c) of claim 1).

3.8 In D8, when all tasks have reported data for a frame, a number of different statistics are computed for the completed frame, including the performance of individual tasks and that of the pipeline as a whole on that frame. The real-time instrumentation system (SPI) maintains a repository of monitored data for use in the detection and allocation decisions (section 4.2.1 "Monitoring", second and third paragraphs).

3.8.1 The method of D8 thus includes the step of "collecting said information signals" (which the board will itemise as feature (d) of claim 1).

3.9 The appellant has argued that the data sent to the SPI related to performance of the sub-task of the pipeline stage in the application and so represented application-specific performance measures and not measures related to the operating status of resources under a processing (testing) workload (statements of grounds, point 3.10).

3.9.1 The board considers that the real-time instrumentation system (SPI) of D8 makes it possible to continually/dynamically update the execution profiles of individual tasks to capture their performance as a function of the resources allocated to them. An "execution profile" of a task describes "how a task's performance changes with the number of processors allocated to it" (section 4.4.1 "Processor allocation"). In this manner, the decisions on allocation of resource(s) or processors to the stages of each pipeline of an application can be based on the most recent behaviour of the application (section 4.2.1 "Monitoring", third paragraph). Computation of a new allocation of resources is performed by (incremental) allocation algorithms when a reallocation trigger is received (section 4.4 "Resource allocation and assignment", first sentence; section 4.4.1 "Processor allocation"; section 4.4.2 "Assignment", first paragraph).

It follows that, in D8, the performances of each task represent measures relating to the number of processors operating on that task, and thus to the "operating status" of each of the processors, under the processing workload of the (complete) application.

3.10 The method of D8 thus includes the step of "providing a rule engine and selectively modifying, as a function of the rules in said rule engine and said information signals collected, at least one of:

- said processing workload on said application, and

- the operating status of at least one processing unit in said grid" (which the board will itemise as feature (e) of claim 1; in D8, the "operating status of at least one processing unit in said grid" is modified).

In particular, D8 modifies the operating status of at least one processing unit in said distributed computer architecture/grid since it allocates processors to the stages of each pipeline or to individual tasks.

3.11 The board does not agree with the examining division that the subject-matter of claim 1 differs from the method of D8 in that the modification of at least one of said processing workload on said application and the operating status of at least one processing unit would not be performed by a rule engine as a function of the rules in said rule engine (and said information signals collected) (see decision, point 2.2).

The application discloses that "resource occupation policies" are the capacity for the distributed-component application to ensure that the hardware resources on the remote servers 12 are occupied on the basis of predefined rules (page 17, lines 30 to 34).

The board considers that the (pre-defined) detection/reallocation triggers and (pre-defined) allocation algorithms (Figure 3 of D8; section 4.3 "Detection", last two paragraphs) of the method of D8 can be seen as (pre-defined) "rules" of a "rule engine".

3.12 Therefore, the subject-matter of claim 1 of the main request is not novel over document D8 (Article 54 EPC).

B. First auxiliary request

4. Inventive step over D8

4.1 In the method of D8, as every sub-task of a pipeline stage in the application completes its work on a frame, it sends data about its performance on that frame to the real-time instrumentation system. This data consists of the frame identifier, task and sub-task identifiers, and the total time spent by the sub-task in computation and intra-task communication. When all tasks have reported data for a frame, a number of different statistics are computed for the completed frame, including the performance of individual tasks and that of the pipeline as a whole on that frame (section 4.2.1 "Monitoring", second paragraph).

Thus, in D8 the test procedure is completed, at a frame level, "when all tasks have reported data for a frame" as a "final operating condition associated with said [deleted: test procedure] frame" being reached.

It is obvious that as soon as all frames will have been processed the test procedure of the method of D8 will terminate. In any case, the board considers that feature (i) was notorious for testing or monitoring systems. Thus, feature (i) appears obvious.

4.2 Feature (h) appears to further define the "load testing" of feature (f) and (f'). However, this is indeed the interpretation by the board above, in relation to claim 1 of the main request (see points 3.6, 3.6.1 and 3.6.2, among others). Therefore, applying the analysis of claim 1 of the main request, the unqualified mention in feature (h) of claim 1 of the first auxiliary request of "a number of workflows to be executed" and the fact that "each workflow comprises a sequence of jobs that are to be executed" seem to be obvious features. These obvious features are the steps of selecting a test procedure, which comprise feature (g).

4.3 Thus, the subject-matter of claim 1 of the first auxiliary request does not involve an inventive step over the method of D8 (Article 56 EPC).

C. Second auxiliary request

5. Admissibility

5.1 The second auxiliary request was not admitted into the proceedings by the examining division. The reasons given in its decision by the examining division were:

i) at the time of filing, no reasons were given as to why said amendments had been made and how they intended to overcome the inventive step objection raised in the summons and (ii) it was not immediately (prima facie) apparent to the examining division, even after a discussion with the appellant's representative during the oral proceedings, how amended claim 1 of the second auxiliary request constituted a "convergent development" of the subject-matter of claim 1 of the main or first auxiliary request, as the added features were not directed either to further specifying details of the "rule engine" nor to further specifying details on how/when to "terminate the test procedure" (both of which were considered by the examining division as distinguishing features of claim 1 of the main and first auxiliary requests over the method of D8). On the contrary, said added features related to adapting the processing workload during the test procedure based on input relating to the operating status of the processing units and prima facie addressed a different technical problem from those addressed in the main and the first auxiliary requests. They would therefore "give rise to an additional (third) inventive step objection (which was already present in the summons, par. 5.2)" (see also claim 2 as originally filed). For these reasons the examining division considered that the "clear allowability" criterion, as defined in the Guidelines for Examination, H-II, 2.7.1, was not met, and decided not to admit the second auxiliary request in the proceedings.

5.2 In the letter of 29 June 2017, the applicant indicated only the basis for the amendments for this second auxiliary request and made no submission in substance about the aim of this request with regard to inventive step. In the letter of reply to the board's communication, it argued that claim 1 of the second auxiliary request was limited, since it further specified "how to selectively modify the processing workload on the application based on the processing of the information signals indicative of the operating status of the processing units" (point 4.1). But in the grounds of appeal (see point 5.28), too, the appellant did not indicate how the amendments would overcome the inventive step objections (in view of document D8 or D8 in combination with document D2; see point 10.1 of the decision under appeal for the inventive step objection in relation to claim 2 of the main request, which was inserted into previous claim 1 of the first auxiliary request).

5.3 The board has some doubt as to whether the late-filed second auxiliary request represents a convergent development of the first auxiliary request in view of the examining division's arguments.

5.4 Therefore, the board decides to exercise its discretion under Article 12(4) RPBA 2007 to hold inadmissible requests which were not admitted in the first-instance proceedings. In this case, the board does not see any indication that the examining division exercised its discretion of not admitting the second auxiliary request on the basis of the wrong principles or in an unreasonable way (this appears also to be acknowledged by the appellant in its letter of reply dated 16 April 2021, point 1.9).

C. Third auxiliary request

6. Admissibility

6.1 The third auxiliary request was not admitted into the proceedings by the examining division. The reasons given in its decision by the examining division were that the features added in the third auxiliary request related to: (i) adapting the workload during the test procedure based on input relating to the operating status of the processing units; and (ii) defining how the sensing and adaptation of the operating status of the processing nodes on the grid is performed. It prima facie addressed two further technical problems, different from those addressed in the main and the first auxiliary requests, and would therefore "give rise to additional (third and fourth) inventive step objections (which were already present in the summons, par. 5.2-5.4)" (see also claims 3 to 5 as originally filed).

6.2 In its letter of 29 June 2017, the applicant indicated only the basis for the amendments for this third auxiliary request and said nothing about the aim of this request with regard to inventive step. Nor did the appellant indicate in the grounds of appeal (see point 5.28) how the amendments would overcome the inventive step objections (in view of document D8 or D8 in combination with documents D1 and D4, see points 10.2 and 10.3 of the decision under appeal for the inventive step objection in relation to claims 3 to 5 of the main request which were inserted into previous claim 1 of the second auxiliary request). In the letter of reply to the board's communication, it only argued that claim 1 of the third auxiliary request was limited further, since it specified how the information signals were collected and how the operating status of the at least one processing unit in the grid was modified" (see point 5.1).

6.3 The board has some doubt as to whether the late-filed third auxiliary request represents a convergent development of the second auxiliary request in view of the examining division's arguments.

6.4 Therefore, the board decides to exercise its discretion under Article 12(4) RPBA 2007 to hold inadmissible requests which were not admitted in the first-instance proceedings. The board does not see any indication that the examining division exercised its discretion of not admitting the third auxiliary request on the basis of the wrong principles or in an unreasonable way.

7. Alleged procedural violation and request for remittal

7.1 In the statement of grounds, the appellant presented a history of the case. D8 was cited for the first time by the examining division in the summons, dated 3 March 2017 to attend oral proceedings scheduled for 7 July 2017. The summons was issued eight years after the appellant's reply to the communication under Article 94(3) EPC had been filed. In this summons, there was an objection to independent claim 1, as lacking an inventive step over D8. On 29 May 2017, the examining division rejected the appellant's request of 11 May 2017 for postponement of the oral proceedings. The request had been prompted by the appointment of a new representative on that date. The appellant argued that the representative would not have had enough time to study the case and prepare for the oral proceedings after taking over the representation. The examining division also rejected ,a request (made during a telephone consultation on 31 May 2017 with the first examiner of the case, for which minutes were sent on 6 June 2017) to continue the examination in writing.

7.2 On 1 June 2017, the appellant filed a main request with an unamended claim 1, together with arguments. On 22 June 2017, the examining division informed the appellant that it was not convinced by the latest arguments. On 29 June 2017, the appellant filed three auxiliary requests; this was later than the final date for making written submissions and/or amendments (which was set at 7 June 2017). According to the appellant, the lateness of these requests was due to the time necessary for the applicant's representative "to contact the Applicant and agree on the claim amendments".

7.3 During the oral proceedings, an appellant's further request to continue the examination in writing was rejected.

7.4 The appellant is of the opinion that summoning it to oral proceedings (following a conditional request by the appellant itself) was not the appropriate means of dealing with a new substantive objection based on a new prior-art document prejudicing the grant of the patent application (statement of grounds, point 4.10).

7.5 Furthermore, the appellant argued that the choice of the examining division to summon the appellant to oral proceedings instead of issuing a written communication under Article 94(3) EPC clearly penalised the appellant, who was thereby granted a period of time to prepare a reply to a new objection that was shorter than the minimum period typically granted in examination proceedings to reply to a communication under Article 94(3) EPC (nearly three months, rather than four to sixth months) (statement of grounds, point 4.14).

The board notes that the examining division has some discretion in deciding when to issue a summons to oral proceedings. In addition, it is the responsibility of the appellant to decide when to change its representative. In this case, the representative was changed after the summons had been received and before the appellant's submission of 1 June 2017 (see letter of reply dated 16 April 2021, point 1.6 and statement of grounds, point 1.8).

7.6 Referring to the Guidelines for Examination, C-IV, 7.2, the appellant argued that the examining division did not provide any comment about the required "exceptional circumstances" according to which the examiner had not been barred from looking for a new document "at the very last stage of examination". This would constitute a fundamental deficiency of the first-instance proceedings (statement of grounds, point 4.13).

7.7 The board does not agree with the appellant that the course of action in the first-instance proceedings constituted (a) fundamental procedural violation(s).

Exceptionally, when an examiner believes that material relevant to obviousness might be found in technical fields not taken into account during the search, an additional search during examination may be necessary (Guidelines for Examination, C-IV, 7.2, point (vii)). The board is not aware why document D8 was cited for the first time in the summons to oral proceedings before the examining division; but although in principle all search work should be done at the search stage, examiners are not barred from looking for a relevant document whose existence they know of or have reason to suspect, if they can retrieve that document "in a short time" (Guidelines for Examination, C-IV, 7.3).

The appellant argued that the "eight-years period of the Examiner's inactivity on this case" did not satisfy the requirement of a "short time" (see letter of reply dated 16 April 2021, point 1.2). However, the board points out that the expression "in a short time" refers here to a period from which an examiner recommences work on the particular case, since an examiner works on a plurality of cases and may interrupt the examination of one case when working on another. The board notes that the appellant did not enquire of the EPO, during this eight-year period, about the date when it could expect a further communication from the examining division, even though the appellant might have expected it to be a proposal for grant under Rule 71(3) EPC or a "communication with minor issues" (see letter of reply dated 16 April 2021, point 1.5). Nor did the appellant make a request for accelerated examination.

7.8 The appellant complained that "if the Applicant would have withdrawn his request for oral proceedings after receiving the summons, the Applicant would have had no opportunity to comment on the newly raised objection [note from the board: based on document D8]. Besides, it has been quite surprising reading [in the communication accompanying the summons to oral proceedings] about the possibility to request a decision according to the state of the file at this point, as the file did not contain any comments by the Applicant about the inventive step objection prejudicing the grant of the patent.". The appellant was of the opinion that the statements made by the examining division in said communication clearly contravened Article 113(1) EPC (statement of grounds, points 4.11 and 4.12).

However, the board is of the opinion that it is the sole responsibility of the appellant whether it withdraws a request for oral proceedings or requests a decision "according to the state of the file".

7.9 Furthermore, the board observes that in its letter of 1 June 2017 the appellant (applicant) did not object to the introduction of document D8 into the examining procedure, but merely mentioned the fact that this document was cited for the first time with the summons.

In its letter of 1 June 2017 (see page 2), the appellant (applicant) mentioned that it "has carefully considered the Examiner's objection" and that "detailed arguments are provided herein below" (see pages 2 to 4).

At that stage of the procedure, no objections about an excessively short time limit or any violation of the right to be heard were raised by the appellant. On the contrary, in a further letter of 29 June 2017 the appellant (applicant) filed amended claims but did also not raise any further objection as to the conduct of the examining procedure.

According to the minutes of the oral proceedings before the examining division it was only at that later stage that the appellant felt compelled to raise procedural objections (see point 5 of the minutes).

Nevertheless, it appears that the appellant had sufficient time to address all the objections made by the examining division, including those relating to document D8, and had the opportunity to amend its case by filing further auxiliary requests.

Therefore the board is of the opinion that no fundamental procedural violations occurred and that the request for remittal to the examining division for these reasons, even if a new examining division were appointed, is not allowable as no special reasons for remitting present themselves (see also Article 11 RPBA 2020).

Consequently, the request for reimbursement of the appeal fee is rejected.

8. As none of the appellant's requests can form the basis for the grant of a patent, the appeal is to be dismissed.