wiki:history
Last modified 8 years ago Last modified on 11/09/11 10:11:27

History

The EARTHWORM project was started in 1993, mainly in response to a number of common needs identified by regional seismic networks: The automatic processing systems used by most regional networks were aging, and maintenance costs were growing; advances in seismic research required data from new, sophisticated sensors in order to provide viable research data; growing demands for immediate societal utility called for new and highly visible real-time products; and finally, shrinking funding meant that most individual networks could no longer support local system development efforts.

Most of the processing systems then in use had a number of drawbacks which made it difficult to enhance them to meet the new requirements: They were built on vendor-specific products, which tied them to high-price vendors, and deprived them of the benefits of the rapidly growing mass market. They tended to be tightly integrated, in the sense that functionally unrelated parts of the system shared a common resource, such as processor cycles, internal data paths, or peripheral devices. This resulted in systems that were difficult to modify, as changes to one function could cause malfunctions in an unrelated area.

The initial EARTHWORM effort had the specific mission of replacing the aging Rex Allen / Jim Ellis Real Time Pickers (RTPs) at Menlo Park. These devices had functioned well for decades and had become the mainstay of Menlo Park's ability to produce rapid event solutions. However, replacement parts were no longer available, and there was an urgent need to produce a replacement system in minimum time. This meant producing a system which could process over five hundred channels of data quickly and reliably.

In response to this, a delegation from Menlo Park was dispatched to present these problems to Carl Johnson, then at the University of Hawaii. Carl had produced many, if not most, of the seismic processing systems to date. The hope was that having been instrumental in creating the current situation, he might be in position to assist us in getting out of it. The visit bore fruit. Carl provided a number of ideas and algorithms that became the core of a new, common real-time seismic processing system.

Since the initial objective of this system was to provide rapid notification of seismic events, the system that evolved had no 'memory' of past events. The emphasis was on speed and reliability, and an event was completed when the system produced a notification. As the project progressed, however, we were requested to address additional needs, such as interactive review of acquired events, association of late-arriving data, incorporation of strong motion data, and production of catalogs and archive volumes. This interactive earthworm as initially part of early 6.x versions and then abandoned. Earthworm currently (as of 2011) has no interactive reveiw.

Design goals

One of Carl's most significant contributions is a set of principles to guide the design and implementation of a seismic processing system that would avoid some of the pitfalls of its predecessors:

Modularity: each function performed by the system should be encapsulated into a module that can function independently of other modules, in terms of hardware as well as software. The implication is that a set of critical system functions can be guaranteed to be independent of other functions in the system. Thus, new, experimental functions can be added without disrupting pre-existing critical operations. Also, quality assurance can be performed on a module-by-module level, rather than for the entire system. This has proved to be a difficult principle to adhere to in practice. First, one has to resist the temptation to use intermediate results already computed by other modules, and second, to resist the lure of using various common utilities which would streamline operations. System independence: Such modules should operate on various brands of computer hardware and operating systems, and various types of such computers can be linked together to operate as one system. This, with the idea of modularity above, implies that the system can gradually migrate from one type of computer to another without disruption. In practice this means using only standardized portions of various computer systems, and isolating any unavoidable system- specific functions in standardized wrapper routines. Scalability: The system is to provide a smooth cost-performance curve to accommodate large as well as small networks. This was perhaps more significant prior to the availability of cheap, powerful mass-produced computers. However, it is still relevant in regard to license fees for included commercial software. Connectivity: The assumption is that such systems are no longer isolated, but have to interact quickly and reliably with other automatic real-time systems, interactive analysis systems, and various notification schemes. The objective is to provide connectivity at various levels of automatic and interactive processing, so that such systems can be configured to operate in configurations ranging from complete stand-alone operation to functioning as nodes in a distributed wide-area system. Robustness: Traditionally, the role of seismic processing systems was mainly one of research support. As a result, system reliability was naturally less important than cost, time to completion, and features. However, the new responsibilities to the press and emergency response agencies required very high levels of system reliability, especially during seismic crises, when input data and power may be interrupted, and system loads may increase dramatically. Thus issues like error detection and recovery, time-to-repair, graceful degradation, and load control were becoming vital, and had to be designed into the system. The costs of robustness are often not appreciated, in terms of time to design, implement, and test, as well as costs for suitable hardware and ancillary equipment.

System Architecture

Carl Johnson also provided a system architecture which implemented the above design goals. This basic scheme was developed by the EARTHWORM team at Menlo Park, and resulted in the scheme described below.

An EARTHWORM system consists of a set of modules 'immersed' in a message passing 'medium'. Each module performs some coherent task, such as data acquisition, phase picking, etc. Modules communicate by broadcasting and receiving various messages such as packets of trace data, phase picks, etc. The message passing scheme is analogous to radio communications: It consists of a message-carrying 'medium' and a set of standard routines which can be thought of as multi-frequency two-way radios operating in this medium. Modules can use these routines to broadcast messages into this medium on a specified 'frequency', and to 'tune in' to messages on specified 'frequencies'. Significant properties of this scheme are that: (1) a module cannot be prevented from broadcasting a message whenever it chooses; (2) any number of modules can listen in to some sending module(s) without affecting that module; (3) a module receives only the messages of interest to it and it is notified if it missed a message.

This 'medium' (the EARTHWORM transport layer) operates within one computer as well as between different computers. Standard TCP/IP network broadcasts (UDP) are used to carry a batch of messages between computers and shared memory regions are used to simulate such broadcasts within a computer. Each participating computer can be equipped with multiple network ports and multiple shared memory regions. Special adapter modules are used to link the message flows between a shared memory region and a network cable. Thus an EARTHWORM system can be configured to have a geometry of message passing paths tailored for a specific application, operating on any number of computers. Note that this scheme operates within one EARTHWORM system, spanning a maximum distance of several hundred meters. Message exchange between remote EARTHWORMs is handled by a different mechanism. The selective 'frequencies' of this 'medium' are implemented by attaching 'shipping tags' to all messages. They specify the module which created the message, the type of message it is, and the installation where the EARTHWORM system is running. Thus, for example, a message may be labeled as being a p-pick produced by the Rex Allen picker at Menlo Park, or a location produced by Hypo-inverse at Seattle. These tags are then used by the receiving routines to provide only the desired messages to a listening module.

The modules are independent main programs. Each module has free use of the file system and other system facilities. However, the EARTHWORM system offers several suites of support routines that can be used by the program to make it a compliant EARTHWORM module. One such suite is the collection of message passing routines (transport routines). These are used to acquire input data by requesting messages of certain type(s), and to broadcast output messages. Another set of routines is used to effect operating system independence by providing a common set of 'wrapper' routines for system-specific functions. For example, the system routines for starting a thread are different on Solaris and NT; the EARTHWORM system provides routines for both systems with the same name and arguments. Thus, if a program uses that routine, it can run on either operating system: the proper system-specific version of that routine will be automatically inserted when the program is built. This suite currently contains such 'wrappers' for all common functions. Routines are added as required. Moving EARTHWORM to a new operating system involves creating a new set of such 'wrapper' routines. EARTHWORM currently offers 'wrappers' for OS/2, Solaris, and Windows NT.

Modules generally read a configuration file at startup time. This file contains basic EARTHWORM parameters which define the module, specify which message 'medium' (message ring) it wishes to listen to, and which message ring it wishes to broadcast it's output messages to. In addition, this file can contain arbitrary operating parameters specific to its operation.

Error detection is accomplished via several mechanisms: Any module may broadcast an error message. Such messages are forwarded to a special module (StartStop?, below), which takes appropriate actions based on the nature of the complaint. Each module can also provide a periodic heartbeat; if this heartbeat should cease an error would then be declared and processed. A module may also request to be restarted if its heartbeat should cease. In addition, the EARTHWORM as a whole can produce a heartbeat to an external monitoring device. Such a monitoring device, capable of generating pager messages, is available as part of the EARTHWORM system.