Management Issues in Transparent Optical Networks R. Rejeb, I. Pavlosoglou, M. S. Leeson and R. J. Green School of Engineering, University of Warwick, Coventry, CV4 7AL, UK. ABSTRACT All-Optical Networks (AONs) are emerging as a promising technology for very high data rate communications, flexible switching and broadband application support. They contain only optical components and vary to a large extent from traditionally optical networks. Due to this specific variation, AONs have unique features and requirements in terms of security and Quality of Service (QoS) that require a very targeted approach in terms of network management. This paper discusses management issues and requirements that arise by using components in communication systems. It then proposes a management framework for the realization of an appropriate Network Management System (NMS) that can meet the challenges posed by AONs. Keywords: All-Optical Networks, Security, Network Management System. 1. INTRODUCTION Network management is an indispensable constituent of communication systems since it is responsible for ensuring the efficient, secure and continuous functioning of any network. Specifically, a network management implementation should be capable of handling the configuration, fault, performance, security, accounting and safety in the network. The configuration management deals with the set of function associated with managing orderly changes, including the management of equipment, connection, and adaptation in the network. discovery, as an important function of configuration management, involves maintaining state information about the network and the current status of connections. Connection management deals with setting up connections, keeping track of them, and taking them down when they are not needed anymore [1]. Performance management is responsible for monitoring and managing the various parameters that measure the performance of the network. It is closely tied into fault management, which deals with fault detection and identification, alarm surveillance and reporting in the network. Although performance management is an essential function that ensures QoS guarantees in the network, it is still a major complication in AONs. Security management involves protecting data on the network, achieved using two mechanisms: controlling and authenticating access to the network and data encryption. It should be noted that optical network security could be classified into two separate, yet interrelated, types, namely physical security and semantic security [2]. The former promises to ensure integrity and privacy of information by protecting the network against service disruption and service degradation in the network. The latter, with which security management is concerned, focuses on the protection of information even when an attacker has access to the transmission data channel. security management for AONs, which is of interest for us, is counted as one of the major challenges of performance and fault management. Accounting management involves keeping track of users' utilization of the network and of the various maintenance costs involved in managing the network. Safety management is an additional function for optical networks, which is needed to ensure the limitation imposed for ensuring eye safety [1]. AONs contain only transparent optical components and therefore differ to a large extent from the optical networks currently used. Due to this specific variation, AONs have unique features and requirements in terms of security and QoS that require a very targeted approach in terms of network management. Although the job of network management for AONs is essentially no different from that of managing traditional optical networks, numerous management issues arise, in particular because of network transparency 1 and the peculiar characteristics of AON components. Whilst some of the available control and management mechanisms are applicable to different types of network architectures, many of these are not adequate for AONs and must therefore thoroughly considered and carefully addressed. These specific features thus require more sophisticated techniques and methods for managing and monitoring the network, controlled with an appropriate NMS that can meet and satisfy the challenges posed by AONs. 2. MANAGEMENT ISSUES Network management for optical networks faces additional security challenges that arise by using transparent optical components in communication systems. Key aspects include: 1. Establishing a robust and flexible optical control plane for managing network resources, provisioning and maintaining network connections across multiple control domains. 2. As domains of transparency become larger, physical constraints on connectivity and transmission become increasingly of concern to the optical control plane. 1 A component or device is called X-transparent if it forwards incoming signals from input to output without examining or modifying the X aspect of the signal. For example, all-optical components are electrically transparent.
3. Addressing security issues that arise, in particular, due to the peculiar behavior of fiber transmission medium and transparent optical components. 4. Agreeing expertise techniques and monitoring methods for measurements of optical parameters. 5. Specifying a comprehensive synthesis of optical parameters and attributes that could be implemented in an appropriate management information model for AONs. 6. Designing suitable functions for managing all the specific features and requirements posed by AONs, taking into consideration network and system scalability as well as interoperability between equipment and components from different vendors. One of the main premises of AONs is the establishment of a robust and flexible optical control plane for managing network resources, provisioning and maintaining network connections across multiple control domains. Such a control plane must have the ability to select lightpaths 2 for requested end-to-end connections, assign wavelengths to these lightpaths, and configure the appropriate optical resources in the network. Furthermore, it should be able to provide updates for link state information to reflect which wavelengths are currently being used on each fiber link so that routers and switches may make informed routing decisions. An important issue that arises in this regard is how to address the tradeoff between service quality and resource utilization. This challenge requires different scheduling and sharing mechanisms to maximize resource utilization while QoS is assured. One possible solution is the aggregation of traffic flows to maximize optical throughput and to reduce operational and capital costs, taking into account qualities of optical transmission in addition to protection and restoration schemes to ensure adequate service differentiation and QoS assurance. A flexible control plane offers dynamic provisioning, performance monitoring, plus efficient restoration in the network and most of these functions need to move to the optical domain. For example, connection provisioning should enable a fast automatic setup and teardown of paths across the network allowing dynamic reconfiguration of traffic patterns without conversion to the electrical domain. Another related issue arises from the fact that the implementation of a control plane requires information transfer between control and management entities that are involved in the control process. To achieve this, fast signaling channels need to be in place between network elements to exchange real-time control information to manage all the supported connections and to perform other control functions. In general, control channels can be realized in different ways; one might be implemented in-band while another may be implemented out-of-band. There are, however, compelling reasons for decoupling control channels from their associated data links. An important reason is that data traffic carried in the optical domain is transparently switched to increase the efficiency of the network and there is thus no need for Optical Cross-Connect (OXC) switches to have any understanding of the protocol stacks used for handling the control information. Another reason is that there may not be any active channels available while the data links are in still use, for example when bringing one or more control channels down gracefully for maintenance purposes. From a management point of view, it is unacceptable to tear down a data traffic link, simply because the control channel is no longer available. Moreover, between a pair of core OXCs there may be multiple data links and it is therefore, more efficient to manage these as a bundle using a single separated out-of-band control channel. In recent years, the notion of an optical control plane has received extensive attention and has rapidly developed to a detailed set of protocol standards, currently being standardized by the International Telecommunication Union- Telecommunication Standardization Sector (ITU-T) and others [3]. Nevertheless, several additional issues in terms of security and network management are still unsolved [2]. One of the main management issues revolves around the fact that very little is currently understood concerning performance monitoring and fault management for AONs. When discussing routing in a transparent network, it is usually assumed that all routes have adequate signal quality (ensured by limiting AONs to sub-networks of limited size). This approach is very practical and has been applied to date, for example when determining the maximum length of optical links and spans. Specifically, operational considerations like failure isolation also make limiting the size of domains of transparency very attractive. Efficient connection provisioning, however, requires more than simply advertising the availability of wavelengths and routes to switches and routers in the network. Routing and wavelength assignment (RWA) requires additional information to be taken in account about the characteristics and performance measurements of all supported lightpaths in the network. Thus, the combination of both RWA and performance information, which should be disseminated and shared between network elements, will allow the adequate assessment of the status of all connections simultaneously. In particular, this necessitates enhancing the routing protocols to update and advertise the attributes and parameters that are necessary to compute routes for requested connections and pre-compute the corresponding restoration paths. To satisfy this requirement, transmission impacts associated with the peculiar behavior of various transmission links and other transparent optical components must be not only taken in consideration at the network designing stage, but also continuously, during normal operation because: 1. Transparency in AONs may introduce significant miscellaneous transmission impairments that aggregate 2 A lightpath is defined as end-to-end optical connection between a source and a destination transparent node.
and can impact the signal quality enough to reduce the QoS without precluding all network services. 2. A nefarious user can take advantage of an AON's vulnerabilities to attacks and perform service disruption attacks despite careful network design. 3. An attack, which is erroneously identified as failure, can spread rapidly through the network. 4. Inappropriate action might be taken by the NMS, if both the locations and types of attacks are not identified. However, there are many complications that prevent the continuous control and monitoring of all supported lightpaths simultaneously. An important implication of using transparent optical components in communication systems is that traditional methods that are used to manage and monitor the health of the network may no longer be appropriate. Therefore, without additional control mechanisms 3 a break in the core of an optical network might not be detectable. Another implication revolves around the fact that service disruption and security failure identification of both the location and type may differ significantly from traditional optical networks [4]. In particular, attacks must be detected and identified at any point in the network where they may occur. In an AON, a lightpath will be capable of carrying analogue and digital signals with arbitrary bit rates and protocol formats. Such a lightpath typically traverses multiple data links from source to destination and there are many components along its route that can fail. Since a transmission fiber carries a number of high-bandwidth channels, its failure will cause simultaneous multiple channel failures, resulting in large amounts of data being corrupted or compromised. Moreover, the lack of data traffic regeneration ability at intermediate nodes within AONs also significantly complicates the segment-by-segment testing of communication links, since the transparent components will not by design understand the modulation and coding of the signal. This makes failure detection and fault localization difficult using current integrity test methods [2]. An integral part of any NMS is performance management, since it provides signal quality measurements at very low Bit Error Rates (BERs), and fault diagnostic support for fault management. In particular, performance management is an area of great interest in optical networking because of its essential role in monitoring the various parameters that measure the performance of the network. It is, therefore, an important management component because it is needed to provide inputs and control information to other management functions (especially fault management) when anomalous conditions are detected in the network. Performance management is still a major complication in AONs, since optical performance measurements, which are typically limited to optical power, Optical Signal-to-Noise Ratio (OSNR), and wavelength registration, do not directly relate to QoS measures used by carriers. The main complication of performance management is that carrier QoS measurements are concerned with attributes related to the lightpath, such as BER and parity checks, of which the management system may have no prior knowledge. Moreover, transparency means that it is not possible to access overhead bits in the transmitted data to obtain performance-related measures, adding further complexity to the detection of service disruption. Unless information concerning the type of signal that is being carried on a lightpath is conveyed to the NMS, it will not be able to ascertain whether the measured power levels and OSNR fall within the preset acceptable limits [1]. Another problem related to transparency lies in determining the threshold values for the various parameters at which alarms must be declared. Often an optimum choice of these values can only be made with prior knowledge of several parameters such as the bite rate, received power levels, and different noise components. Since all these parameters depend on attributes of lightpaths already in use, one cannot optimize these threshold values in a static sense. On the other hand, dynamic control of threshold for each lightpath would add a significant communication overhead for the network. As previously stated, fault management refers to the set of functions responsible for detecting transmission failures when they happen, isolating failed components, restoring connectivity services, and alerting the NMS appropriately through alarms and rapid notifications. Fault management functions should be rapid and accurate to avoid substantial data loss or corruption and also must accomplish their task over an extended period of time. Moreover, fault management must be able particularly to detect indirect transmission failures, which can impact the signal quality enough to reduce the QoS without precluding all network services. One of the main complications of fault management is that detection functions, which should be handled at the layer 4 closest to the failure, is delegated to the physical layer instead of higher layers. That is, fault detection and localization methods are less insulated from details of physical layer than of higher layers, requiring the availability of expertise diagnostic techniques to measure and control the smallest granular component, the wavelength channel. Moreover, fast and accurate determination of the various performance measures of a wavelength channel implies that measurements have to be done while leaving it in the optical domain. One possible way to overcome this difficulty is by taking a small part of the optical power from a desired channel via a low loss tap to a measurement device such as a power-averaging receiver. With regard to alarm management, an extensive management effort needs to be supported, in order to prevent the complete service disruption of the entire network. This may occur because a single failure event may cause multiple undesirable alarms that spread quickly though the network and hence cause incorrect action to be taken in response to the failed condition. Since an optical signal is not terminated at an intermediate node, if a 3 Control mechanisms include both hardware and software support 4 Here layers refer to the classical network layered hierarchy proposed by the International Standards Organization (ISO).
wavelength fails, all nodes along the path downstream of the failed wavelength could trigger an unnecessary alarm. This can lead to a large number of alarms for a single failure, and makes it somewhat more complicated to determine the cause of the alarm. Dealing with this challenge requires more sophisticated methods and algorithms that allow reporting the root-cause alarm of the actual failure, and suppressing the remaining alarms. 3. MONITORING METHODS From a management perspective, fault management functions may be broadly classified in three categories, namely prevention, detection, and reaction. As already mentioned, fault detection deals with detecting various hardware and software failures when they happen and with isolating failed components. It encompasses failure event recognition, failure identification, and alarm generation through appropriate notifications. Fault protection and restoration refers to the processes used to protect against failed conditions and to restore network services in the event of transmission failures. In protecting optical networks, supervisory techniques are required, enabling the detection of failures, and these may currently be grouped into two categories [4]. The first is based on methods that perform statistical analysis of the transmitted data: power detection, Optical Spectral Analyzers (OSAs), and BER Testers (BERTs). The second relies on the use of probe signals devoted to diagnostic purposes: pilot tones and Optical Time Domain Reflectometers (OTDRs). However, most of these supervisory methods are insufficient to detect small and sporadic BER degradations, failing to detect in-band and out-ofband jamming attacks. Even the use of probe signals is not sensitive enough to detect the BER degradations. Fault detection capabilities of available supervisory techniques [5, 6] and monitoring methods are summarized in Table 1. Recent proposals to overcome the difficulty of monitoring the continuity and estimating the quality of optical signals include error detecting codes, sampling, and spectral methods [4, 7, 8]. However most of these are too difficult to implement in every transparent optical component or require the access to the electrical domain. Since the key performance parameters associated with a lightpath such as the BER can only be monitored when the signal is available in the electrical domain, some work has been proposed on adapting schemes for digital communications [2, 4, 9] providing new approaches for fault detection, isolation, and recovery in AONs. Table 1: Failure detection capabilities of available supervisory and monitoring methods Supervisory and Monitoring Techniques Signal In-Band Out-Band Time Noise Accuracy Domain Jamming Jamming Distortion BER measurements Digital Yes Yes Yes Yes High Sampling methods Optical Yes Yes Yes Yes Medium Optical Power Meters Optical No No No No Low BER Testers Optical No No No No Low Optical Spectral Analyzers Optical No Yes No No Low Pilot tones Optical No No No No Low Optical Time Domain Reflectometers Optical No No No No Low Optical Frequency Domain Reflectometers Optical Yes Yes No No Medium Regarding cost and implementation plausibility, available monitoring methods may be listed in decreasing order of difficulty (and also decreasing information content) as follows [5]: 1. BER estimation with access to digital signal BER measurement methods. 2. BER estimation with access to optical signal Sampling methods (Eye monitoring). 3. Power and noise monitoring - Optical Signal Noise Ratio (OSNR). 4. Indirect methods are methods that do not directly sense the signal shape or power but still indicate the proper function of network components [5]. The conclusion is that the problems rising from physical security in AONs and the means of protecting against them cannot be tackled using available supervisory methods. Despite new methods for detection and localization of attacks having been proposed, no robust standard or technique exists to date for QoS monitoring in AONs. All proposed techniques and schemes are still in their infancy, offering new directions for future research. 4. MANAGEMENT FRAMEWORK Given the complexity of network management and diversity requirements, a valid question to ask is whether a centralized NMS, would be better than decentralized management methods. Although most of available management systems are implemented in a centralized manner, this manner of implementation is rather slow and inadequate for optical networks because of the large communication overhead involved in the process [1]. Furthermore, it becomes too difficult for a single centralized NMS to manage the entire network when this becomes very large. In contrast, decentralized methods are usually much faster and more efficient than centralized management systems, even in small networks with only a few nodes. This is not only because distributed methods involve low communication overhead, but also because certain management functions require rapid and accurate actions to be taken in response to failed conditions. For example, if rapid rerouting is
required, it is not possible to rely on a centralized management system to compute a recovery route for each failed lightpath on demand after a failure has been detected. Distributed systems would clearly not have this difficulty. Network management is performed in a hierarchical manner involving multiple management systems, which are usually implemented in master-slave relationship between managers and associated agents. The key concept applied in this manner of implementation is that of a layering architecture, illustrated in Figure 1(a). Layering refers to the ability of a network to nest finer-granularity over coarser-granularity by hiding details and complexities in different management levels. Thus, these levels would advertise only a standardized abstraction of their state information. In such a layered model, the management functions with associated information can be decomposed into number layers of abstractions. At the border between two adjacent layers, for example Layer n- 1 and Layer n, the management view is presented to Layer n-1. It is presented in the form of management information that is contained within the agent at Layer n. For security and other reasons, the management view that is presented to a higher layer need not unveil all details of a low layer. An agent at an arbitrary layer is then responsible for providing the management information that is necessary at the immediately higher layer. Since the principle of layering can be applied in a recursive fashion, the management view of Layer n+1 can be presented to Layer n, and so on. Layer n-1 Manager NMS Network Management Layer Layer n Manager EMS EMS Element Management Layer Layer n+1 Lightpath P i P i+2 P j Detection points P j+3 P k+1 (a) (b) Figure 1: Management and functional architecture in a typical optical network In such a hierarchical model, the individual components to be managed are referred to as network elements. For AONs, these include optical line terminals, optical add/drop multiplexers, optical amplifiers and OXCs. Each network element is managed by its Element Management System (EMS) and is represented by an information model that contains a variety of attributes associated with it and a set of control and measurements parameters for monitoring purposes. The network element has a built-in agent that is responsible for performing control functions such as monitoring and reporting any performance degradation of wavelength channels. As shown in Figure 1(b), the EMS is usually connected to one or more agents and communicates with neighboring EMSs using separated control channels. Hence, the EMS typically has a view only of the network elements that it can manage and may not have a comprehensive view of the entire network [1]. As discussed earlier, one of the main management issues in terms of security revolves around the fact that service disruption attacks can spread rapidly through the network. One possible way to address this challenge is by enabling network elements to detect and identify failures or attacks at any point within the element where these may occur. This usually requires monitoring of gradual performance degradation for all connections supported in the network and taking rapid and accurate actions in response to failed conditions. To do so, it is necessary to generate a detailed picture of the state information of various links in a manner that can be communicated to EMSs and other control entities that are involved in the management process. By monitoring the various performance parameters it is possible to estimate the BER, detect gradual or sudden performance degradation, and report this to local or global management systems. As a result, all the available information can be correlated across the network and hence may be used to make decisions regarding fault isolation and to take appropriate actions such as the computing and rerouting of lightpaths. EMSs should also be enabled to archive statistics and historical information by local saving of performance measurements that can be continuously used for connection provisioning and making appropriate decisions as wavelength channels are dynamically managed. Furthermore, this control information may be very useful in analyzing the occurrence of anomalous conditions and service disruptive attacks. This is particularly true for localizing and tracing back of subtle forms of transmission impairments and attacks, which may cause cumulative QoS degradation as optical signals travel through the network. To satisfy this requirement, control information should be expanded to include additional performance measurements, which will be carried and shared between network elements. To achieve this, a comprehensive synthesis of optical attributes is necessary to distinguish the parameters that change frequently, requiring continuous exchanges between network elements, from those that change infrequently and often depend on the characteristics of the network element itself [10]. The resulting specification can be used to adapt and expand existing protocols and information models. As illustrated in Figure 1(b), a lightpath can be modeled as a sequence of detection points [9]. An agent therefore, may be able to manage one or more detection points, simultaneously. Thus, there is no need to handle P k
Mana gement Mana gement Mana gement every detection point individually. For example, an OXC node, which is composed of several active and passive optical components, can be considered as a lightpath segment or a subset of several detection points that may be managed by one or more agents. Accordingly, the corresponding EMS may communicate to certain agents to examine or alter some information, without the difficulty of connecting to each detection point individually. It will in fact need to collect a view of performance information, which is only part of the information illuminated at each detection point. In short, the agent may act as a proxy for certain EMSs hosted elsewhere which only need to contact the agent to get the information they need. In conclusion, additional management functions need to be implemented at the agent and EMS components of each network element. Special attention needs to be paid by analyzing and specifying interfaces, protocols and the information content that should be exchanged between EMS-EMS and EMS-agent. 5. NETWORK ARCHITECTURE The design of an optical network is an important and very practical issue. As stated above, a desirable architecture should feature, inter alia, flexible management, automatic lightpath protection and restoration, and the ability to compile an inventory. Moreover, network architectures should support the gradual introduction of new technologies into the network without time consuming and costly changes to embedded technologies. However, network architectures currently used may be categorized in two main models, namely the overlay model and the peer model. Although both models consist essentially of an optical core that provides wavelength services to client components 5, which reside at the edges of the network, they are intrinsically different and offer up two key concepts for provisioning and managing traffic flows in the network. a) overlay model b) peer model Figure 2: Network level abstraction The overlay model, shown in Figure 2(a) hides the internals of the optical network and thus requires two separate, yet interoperable, control mechanisms for provisioning and managing optical services in the network. One mechanism operates within the core optical network and the other acts as interface between the core and edge components which support lightpaths that are either dynamically signaled across the core optical network or statically provisioned without seeing inside the core s topology. The overlay model therefore imposes additionally control boundaries between the core and edge by effectively hiding the contents of the core. Management Domain D1 NNI Domain D3 NNI Domain D2 NNI (a) Domain management architecture (b) Inter-domain management architecture Figure 3: Management architecture of optical networks The peer model, shown in Figure 2(b), considers the network as a single domain, opening the internals of the core optical network to the edge components making the internal topology visible able to participate in provisioning and routing decisions. Whilst this has the advantage of providing a single control and management system in the network, there are some significant considerations: The availability of topological information to all components in the network makes this model less secure. New standard control mechanisms are required since available proprietary ones cannot be employed. 5 Component is used here for both for hardware devices and software interfaces
Additionally approaches for traffic protection and restoration are required. Another model, known as hybrid model, combines both the overlay and peer approaches, taking advantages from both models and providing more flexibility. In this model, shown in Figure 3, some edge components serve as peers to the core network and share the same instance of a common control mechanism with the core network. Other edge components could have their own control plane (or a separate instance of the control plane used by the core network), and interface with the core network through the User Network Interface (UNI). 6. CONCLUSION AONs contain only transparent optical components and therefore differ to a large extent from the optical networks currently used. Due to this specific variation, AONs have unique features and requirements in terms of security and QoS that require a very targeted approach in terms of network management. Network management for AONs faces additional challenges such as resource allocation, connection provisioning, dynamic routing, performance monitoring, and ensuring QoS guaranties in the network. Performance management is germane to successful AON operation since it provides signal quality measurements at very low bit BER and fault diagnostic support for the fault management, which is still a major complication for AONs. Due to the analogue nature of optical signals, miscellaneous transmission impairments, which must be continuously monitored and identified in the event of transmission failures, aggregate and can impact the signal quality enough to reduce the QoS without precluding all network services. Moreover, vulnerabilities within AONs can be exploited by an attacker to analyze traffic, eavesdrop, degrade QoS, or deny service altogether. We have discussed management issues and requirements for AONs with particular emphasis on complications that arise due the peculiar behavior of the fiber transmission medium and various optical components. Subsequently we have presented a management framework for the realization of an appropriate NMS that could meet the challenges posed by AONs. In particular, we proposed to expand control information, which should be carried and shared between network elements, to include additional performance measurements that would allow EMSs to assess the health of all connections supported in the network and thus ensure that the desired QoS guaranties are met. REFERENCES [1] R. Ramaswami and K. N. Sivarajan, Optical Networks, A Practical Perspective, Academic Press 2001. [2] M. Medard, D. Marquis, R. A. Barry and Steven G. Finn. "Security Issues in All-Optical Networks", IEEE Network, vol. 3, no. 11, pp. 42-48, 1997. [3] D. Saha, B. Rajagopalan, and G. Berstein, The Optical Network Control Plane: State of the Standards and Deployment, IEEE Optical Communications, vol.1, no. 3, August 2003. [4] M. Medard, S. R. Chinn, and P. Saengudomlert. "Node wrappers for QoS monitoring in transparent optical nodes", Journal of High Speed Networks, vol. 10, no. 4, pp. 247-268, 2001. [5] C. P. Larsen and P. O. Andersson, Signal Quality Monitoring in Optical Networks, Optical Networks Magazine, vol. 1, no. 4, pp. 17-23, Oct. 2000. [6] C. Mas, I. Tomkos and O. K. Tonguz, Optical networks security: a failure management framework, ITCom, Optical Communications & Multimedia Networks, Orlando, Florida, 7-11 September 2003. [7] L. E. Nelson, S. T. Cundiff and C. R. Giles, Optical Monitoring using Data Correlation for WDM Systems, IEEE Photonics Technology Letters, vol. 10, no. 7, pp.1030-1032, July 1998. [8] A. Amrani, et al., Performance Monitor for All-Optical Networks Based on Homodyne Spectroscopy, IEEE Photonics Technology Letters, vol. 12, no. 11, pp. 1564 1566, Nov. 2000. [9] R. Rejeb, I. Pavlosoglou, M. S. Leeson and R. J. Green, Securing All-Optical Networks, ICTON 2003, vol. 1, pp. 87-90, Warsaw, June/July 2003. [10] J. Strand, A. L. Chiu, and R. Tkach, Issues For Routing In the Optical Layer, IEEE Communications Magazine, pp. 81-87, February 2001.