Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2), May 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....be created easily by composing a set of micro generators to suit the needs of a specific application. Such a modular architecture was built on previous work in micro kernel designs and group communication systems. x Kernel, for example, defines a modular structure for implementing network software [9]. The authors of [10] used fault containment wrappers to improve the robustness of COTS micro kernels. By verifying certain predicates when a system call is performed, the wrapper detects errors due to corrupted parameters and may optionally perform some corrective actions to restore the system ....

Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....of layers and components, and run time support for the dynamic behavior. The DIPS framework is able to build stacks with different degrees of customizability. There are other systems as well which have explored composition of network services, like for example NetScript ( DFY98] x kernel ([OP92]) However, they do not automate composition. DIPS provides the needed infrastructure support, that comprises: a stack building framework, that contains a special stack builder which is able to interpret at run time a sequence of component descriptions and build a protocol stack from it . a ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2), May 1992

....power consumption and access control loom large in these designs, and routing strategies have an especially important in uence on the system s performance, and ultimate lifetime. Borrowing design elements from other network simulators, such as WiPPET [19] GloMoSim [7] SSFNet [13] and x Kernel [26], SWAN uses the ISO OSI stack model and provides a standard abstraction for building, at each sensor station, a protocol graph, which consists of a stack of protocol sessions each modeled as a stand alone protocol module. The design has its advantage in that each protocol module is insulated ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110-43, May 1992.

....for a listing of other active network projects. Some non active networking projects share implementation ideas with PLANet. The Fox project [4, 5] implements the TCP IP protocol suite for Ethernet and ATM link layers using the ML programming language, in the approach of the x kernel [17]. This work provides a number of insights into the challenge of implementing network software using a high level language like ML. Their implementation uses SML NJ rather than OCaml so there are some differences; in particular, our programs are byte code interpreted and we make use of OCaml s ....

Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2), May 1992.

....Node Model and Environment Model. 49 An instance of Node Model represents a mobile station. There can be as many instances as the number of mobile stations in the wireless network. Borrowing the design from other network simulators, such as WiPPET [53] GloMoSim [7] SSFNet [29] and x Kernel [94], the Node Model provides a standard abstraction for building protocol graph, which consists of a stack of protocol sessions each modeled as a stand alone protocol module. The design has its advantage in that these modules are insulated from dependence of a particular simulator. The modelers can ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110-43, May 1992.

....of Information This document is intended as a reference manual for a user that is already familiar with the x kernel. There are several other sources of information that you should look at to learn more about the x kernel. First, the x kernel was originally described in a pair of research papers [2, 5]. These are a good place to start to understand the motivation and design rationale behind the x kernel. Second, this Programmer s Manual, while thorough, is somewhat cryptic. It does not serve as a tutorial that teaches you how to write x kernel protocols. For help in learning how to write ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....cannot be re used. For example, the tree establishment and maintenance portion of the protocol might be useful for media thinning or other applications that need to perform processing hop byhop. The approach we used was to consider the processing of each distinct packet type as a microprotocol [19], with one a flow per packet type. This choice was to some extent motivated by restrictions on UNACK MNACK CSM CANEs EE NPM Bowman NodeOS SPM Fig. 5. AER NCA a flows. the capabilities of individual threads within an a flow in Bowman. However, it does leverage the demultiplexing capability ....

S. W. O'Malley and L. L. Peterson. A Dynamic Network Architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....as being capable of dynamically loading new policies at run time. An adaptive system is a flexible system which can, in addition, automatically determine which policy should be used. Flexible systems have been developed in many domains. Flexible group communication systems, such as x kernel [12] and Horus [17] split protocols into elementary modules that can be composed together to obtain the required features. Flexible replication systems allow one to choose which replication policy should be used [3, 8] The 8 same principle has been applied for building routers [9] network traffic ....

Sean W. O'Malley and Larry L. Peterson, A dynamic network architecture, ACM Transactions on Computer Systems 10 (1992), no. 2, 110--143.

....management of multiple distinct resources processors, threads, messages 1 and buffers. One of our primary research goals is to evaluate the benefits of processor cache affinity scheduling in multiprocessor networking 2 . In our experimental environment (consisting of a parallel x kernel [33, 52] running in user space on an 8 processor MIPS R4400 based SGI Challenge XL) packet execution times can vary by as much as a factor of four, depending on the state of the processor cache. This suggests that affinity based scheduling techniques (which have been explored in the context of general ....

....and show that it offers the best overall performance yielding highmessage throughput, high intra stream scalability, and robustness in the presence of bursty arrivals. 6 To establish these results, we began with an unparallelized version of the x kernel protocol development environment [33, 52] running in user space above IRIX 5.2 on our 8 processor MIPS R4400 based SGI Challenge XL. We parallelized the fast path of the UDP, IP, and FDDI protocols in this testbed environment, enabling concurrency both at the packet and connection levels. An analysis of the execution path in UDP IP FDDI ....

Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....Microprotocols Based on the descriptions of the previous section, it is not apparent how the MARS recovery algorithms can be implemented through modular composition or extensions to atomic recovery. To this end, we have applied the concept of microprotocols introduced in the x kernel project [7] to define a set of recovery microprotocols that are decompositions of the recovery algorithms which can be combined to implement various recovery functionalities. As an example, transaction states analysis functionality can be implemented through combinations of the following four simple ....

Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transaction on Computer Systems, 10(2):110--143, May 1992.

....execute in a standard environment such as SunOS or Mach[A 86] GTS(section 2.2.4) Horus(section 2.2.5) and Adaptive(section 2.2.8) are in this category. These systems would either execute at user level or as a server on top of a microkernel such as Mach. 2.2. 1 x Kernel The x Kernel [HP91, WP92] stems from an interest in how the structure of an operating system influences protocol performance. The work has produced a kernel which provides a highly structured environment in which to implement protocols. It is by forcing protocols to be implemented within this defined structure that the ....

....connects to ETH, the Ethernet object, as shown in figure 2.1 CHAPTER 2. SURVEY 10 Data Path TELNET TCP IP Ethernet UDP RPC Figure 2.1: x Kernel protocol graph Protocols are heavyweight objects. For example the whole of TCP or Psync[PBS89] would be implemented in a single object. O Malley[WP92] introduces the idea of lightweight objects within the x Kernel where individual functions of the protocols, such as fragmentation or reliability, are implemented within separate objects. This idea has been incorporated further into the x Kernel and is discussed in section 2.2.2. A Session ....

[Article contains additional citation context not shown here]

Sean W.O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....to meet the needs of a given application [HS93] This is done by configuring the system with just those modules that are needed to provide the functionality required by the application. This approach has also been used to dynamically re configure systems to meet the needs of changing applications [OP92] Abstractions are particularly useful for building fault tolerant software. The use of appropriate abstractions to access the services provided by a module limits access to its internal state by other modules. This helps contain the effect of failures; since modules do not share state, errors ....

Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....system generators will emerge as popular tools for large system development, because they offer a significant advance over existing tools. There are already numerous examples of generators in a wide range of domains, such as databases [Bat88] file systems [Hei90, Hei91] network protocols [Hut91, OMa92], and data structures [Sir93] Our study reveals that even 2 though the systems produced by these generators are radically different in functionality, the organizations of the generators themselves are remarkably similar. The goal of our research is to understand existing software system ....

....Each component implements a basic feature offered by some systems in the domain. Selecting different component combinations corresponds to building systems with different sets of features. We have studied several SSGs, including Genesis (domain of databases) Bat88] Avoca (network protocols) [Hut91, Oma92], Ficus (file systems) Hei90, Hei91] Brale (host at sea buoy systems) Wei90] Adage (avionics) Cog93] and Predator (data structures) Sir93] We observed that several key design properties underlie their organizations: 1. For the purposes of this discussion, the term component is defined to ....

Sean O'Malley and Larry Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110-143, May 1992. 133

....First, protocol code is subdivided into layers or modules and organized into a graph structure representing how those modules combine to provide communication services. Second, protocols are developed within programming environments or subsystems (such as BSD [4] Streams[1, 8] or the x Kernel[6, 2]) which provide an overall protocol model and offer organized, consistent access to operating system resources. Third, protocol code is often ported from one subsystem to another rather than developed from scratch. A protocol s implementation is often tightly coupled with the subsystem in which ....

....graphs and provide an interface between the subsystem and the greater system in which it is embedded, adapter protocols exist at any point in the protocol graph where subsystem boundaries are crossed. Adapter protocols are also similar to the virtual protocols defined in the x Kernel research [6]. Both work transparently within a protocol graph. They do not produce or consume protocol headers nor communicate with remote peers. Like virtual protocols, an adapter need not have a remote peer. We built and measured the performance of several different protocol graphs (shown in Figure 4) ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10:110--143, May 1992.

....implemented in end systems on behalf of individual applications. Thus, sits between the network layer and the application (Figure 1) Much of the prior work on high performance communication falls into four categories: ffl General implementation approaches independent of any particular protocol [5, 10, 8, 11] ffl Implementation techniques focused on existing protocols [12, 6] ffl General considerations for new protocol design [4, 1, 7] ffl Specific proposals for new, efficiently implementable protocols [9, 3] The design of represents a mixture of these approaches: it is both a protocol and an ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992. 6

....activities. 3 A Practical Implementation Method 3.1 Overview of Method The first building block of our implementation method is the careful design of the dynamic activity restructuring to facilitate modular, adaptable, and efficient implementation. We achieve this by combining microprotocols [18], open implementation [13] and incremental specialization [21] We have identified a number of key issues that need to be addressed by an implementation of dynamic activity restructuring, such as: how to control an activity instance s execution under the possible influence of restructuring at ....

....activities. On one hand, our implementation method benefits from several system development techniques and principles that have been demonstrated successfully in other fields like operating systems and programming languages. The idea of mi croprotocol decomposition comes from the x kernel work [18] on building flexible network protocols. Specialization was proposed by the Synthetix project [21] and is closely related with partial evaluation [6] in programming language research. Our approach to system extensibility continues to follow the Open Implementation (OI) principle [13] as in [1, ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....of all these protocol frameworks follow rather coarse grained, layered structuring approaches. The advantages of fine grained structuring and modularization higher flexibility and improved re usability without serious performance degradation were first exposed by O Malley and Peterson [27]. DaCaPo [28] and ADAPTIVE [30] demonstrate the higher flexibility of fine grained modularity by featuring dynamic configuration and assembly of protocols with classified requirements. Bhatti [2] overcomes problems of the X Kernel environment to implement finegrained fault tolerance multicast ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....and process interchangeably. 1 affinity based scheduling policies. The hardware platform serving as the basis for the experimental component of our study is an 8 processor SGI Challenge XL running the IRIX 5.2 operating system. Protocols are implemented using the x kernel framework in user space [8, 14]. Previous work on affinity based scheduling [3, 5, 22, 26] has not established a consensus on its efficacy, even reaching seemingly conflicting conclusions (e.g. 26] vs. 22] None of the work has found affinity based scheduling techniques to be of significant benefit to common applications. ....

Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

....by other types. 5. Conclusion and Outlook Interface definition languages are first instances of an evolving class of general protocol configuration languages and architectures that allow for configuring a protocol that is tailored to application specific requirements (e.g. Clar90] Abbo92b] [O Mal92], Hosc93] Schm93] With each protocol configuration, the most frequently executed path may change. In this case, manual optimization for each configuration will be impractical, and heuristics based on control flow analysis similar to the ones used for presentation conversion are required. ....

O'Malley, Sean and Larry Peterson. A Dynamic Network Architecture. ACM Transactions on Computer Systems, 10, 2 (May 92), 110-143.

....the prediction of the most frequently executed path in a protocol can be automated. Automation of protocol optimization is especially important for the recent trend towards flexible, highly modular protocol architectures that allow for protocol configurations that are application specific (e.g. [O Mal92], Hosc93] Schm93] With a configurable protocol architecture, the most frequently executed path changes with each protocol configuration, rendering manual frequency prediction and fast path implementation difficult or even impossible. In this paper, we discuss an approach for automating the ....

O'Malley, Sean and Larry Peterson. A Dynamic Network Architecture. ACM Transactions on Computer Systems, 10, 2 (May 92), 110-143.

....kernel, the set of available modules was fixed. Later versions of the STREAMS system supported the construction of multiplexor trees and the dynamic loading of STREAMS modules into the kernel. The performance of STREAMS is limited by the execution of each layer as a separate thread. The x kernel [OP92] is a C library for building protocol software developed by the University of Arizona and since used in the OSF 1 Unix kernel. Protocols are implemented in terms of layers and sessions : a layer encapsulates protocol functionality and creates a session to encapsulate the state needed for each ....

S.W. O'Malley and L. L. Peterson. A Dynamic Network Architecture. ACM Transactions on Computer Systems, 10(2): 110-143, May 1992.

....commonly used protocols like TCP and IP are complex, with a variety of functionality and many options. Developing and testing such protocols can be dicult and error prone, and the resulting protocols are not particularly exible. These problems have motivated past research on microprotocols [13, 14]. Each micro protocol embodies a single function or option; more complex behavior is achieved by composing these micro protocols. We will present here two micro protocols: one for a packet checksum, and one for fragmentation. These protocols will also serve to show how chunks provide some ....

....attacks, and public key cryptosystems are notoriously slow. Fortunately, all three of these problems can be addressed, as described by the Secure Active Network Environment (SANE) 3] and implemented in our current testbed, PLANet [7] 6 Related Work Micro protocols are used in the x kernel [11, 13]. However, the stress there is on software engineering and code reuse; although their approach enables the relatively simple development of new protocols, they must still be composed statically, whereas micro protocols implemented with PLAN chunks can be dynamically reordered. Ensemble [14] is a ....

Sean W. O'Malley and Larry L. Peterson. A Dynamic Network Architecture. ACM Transactions on Computer Systems, 10(2):110-143, May 1992.

....in this section to be composed on the proxy host. Parsing and code generation are performed only once for all static services, while structuring the services as independent code transformation filters enables them to be stacked according to site specific requirements [Heidemann Popek 94, O Malley Peterson 92] The proxy uses a cache to avoid rewriting code shared between clients and generates an audit trail for the remote administration console. The code for the dynamic service components resides on the central proxy and is distributed to clients on demand. While the implementation details of our ....

S. W. O'Malley and L. L. Peterson. A Dynamic Network Architecture. ACM 216 Transactions on Computer Systems, 10(2):110--143, May 1992.

....which are shared across sessions. The distinction is immaterial here. The sequence of PEs associated with a session is here called a path. While in most architectures every message associated with a session is handled by the same path, this need not always be the case. In the architecture of [12], for example, the PE structure associated with a user session is actually a graph, and different messages can traverse different paths through this subgraph. Use of the term path notwithstanding, our model does not exclude such architectures. The general scenario by which a user obtains ....

....for that host are not correctly updated, other hosts will be unable to establish communication with that host. While current network hosts change protocol architectures no more frequently than they change addresses, recent research suggests that this may not always be the case in the future [12, 18]. The question of accuracy of the DNS entries has been raised in the Internet community. Indeed the Internet host requirements document [1] specifically warns that a host should not rely on the WKS entries to provide accurate information regarding the services available from another host. These ....

S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110--143, May 1992.

No context found.

Sean W. O'Malley and Larry L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2), May 1992.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC