Assessing Service Properties with Regard to a Requested Qos: The Service Metric

Claudia Linnhoff‐Popien, Dirk ThißenFormal 1999 · DOI · OpenAlex

Introduction

In the late 1990s, the expansion of distributed systems necessitated more sophisticated service discovery mechanisms. Within environments like CORBA (Common Object Request Broker Architecture), the standard "trader" role—acting as a directory for service importers and exporters—faced significant limitations in handling Quality of Service (QoS) requirements. This paper proposes a transition from binary matching to a mathematical "service metric" to improve service selection.

The Limitations of Classic Trading

Traditional trading mechanisms rely on "all-or-nothing" logic. This creates two primary issues:

  1. Lack of Precedence: When multiple services meet a request, the trader cannot rank them objectively. The user receives an unordered list, regardless of which service is actually the "best" fit.
  2. The "Near-Miss" Problem: Services that narrowly fail to meet a strict criterion (e.g., slightly higher latency than requested) are discarded entirely, even if they are practically sufficient.

A Metric-Based Approach

The authors introduce a two-step evaluation procedure to quantify the "distance" between a service request and an offer, effectively turning service selection into a mathematical optimization problem.

The enhanced trading process incorporating a computation layer for distance calculation.

First, a Service Property Record Distance ($d_c$) is calculated. Unlike simple values, requests include target values, bounds, and weights. If a target falls within the provider's bounds, the distance is zero; otherwise, a set of rules determines the numerical deviation.

Second, these individual distances are aggregated into a Service Property Vector Distance ($d_v$). By applying analytic metrics—such as Manhattan, Euclidean, or Maximum metrics—the trader collapses the vector into a single scalar value representing the total "distance" from the ideal request. This allows the system to rank offers by proximity to the user's requirements.

Performance and Results

The implementation was tested in a synthetic environment with 20 service offers and up to 50 properties. A key technical optimization involved sorting properties lexically, allowing the trader to evaluate matches in a single pass rather than performing a fresh search for every property.

Comparison of processing time between the classic trader and various metric-based approaches.

As shown in the performance evaluation, the classic trader's performance degrades rapidly as the complexity of the request increases. In contrast, the metric-based approach exhibits near-linear growth in processing time, making it far more scalable for complex QoS scenarios. Furthermore, the use of multiplicative weighting models proved highly effective at distinguishing between high-quality service offers, providing better ranking accuracy than traditional methods.

Conclusion

By replacing binary logic with a distance-based metric, the authors provided a framework for more flexible and scalable service discovery. While the current model assumes requested properties are present in the offer, it lays the groundwork for more nuanced QoS rule engines in open distributed systems.

Summary updated

BibTeX
@inproceedings{linnhoffpopien1999assessing,
  author = {Claudia Linnhoff‐Popien and Dirk Thißen},
  title = {{Assessing Service Properties with Regard to a Requested Qos: The Service Metric}},
  booktitle = {Formal},
  year = {1999},
  doi = {10.1007/978-0-387-35562-7\_21},
}

Loading related content...

Figure