M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Proceedings of the 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....domain are the Packet Monitoring Program (PMP) 3] and HP OpenView [13] Communication Protocols: Event filters have been used to support efficient packet demultiplexing. An event filter examines incoming packets and forwards them to one or more end points based on values of fields in the packet [16, 18, 28]. Some applications of packet classification in communication protocols include: # Packet demultiplexing: In operating systems, packet filters are used as an efficient technique to classify packets and determine the path the packet must follow within an endsystem to route the packet to its ....

....endpoint [1] # User level protocol implementation: In micro kernel operating systems different protocol stacks may coexist together in user space. Using event filters in the kernel, packets are demultiplexed and forwarded to the proper user level protocol stack. The CSPF [18] BPF [16] MPF [28] and PathFinder [1] are examples of event filtering mechanisms in the communication protocol domain. Active Databases: Active databases are constructed via triggers in a database environment. Triggers are specified as event condition action tuples. When an event occurs and a condition is ....

[Article contains additional citation context not shown here]

Masanobu Yuhara, Brain Bershad, Chirs Maeda, and J. Elliot B. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Winter 1994.

....refine the filtering model. The most widely used packet filter is BPF [18] which uses a register model instruction set, unlike the stack machine model used by [19] Each filter is run on every incoming packet which imposes high overhead on user level programs using it. The Mach Packet Filter (MPF) [24] enhances the BPF model by demultiplexing filter specifications to recognize when two filters use similar patterns. If MPF detects similarities, it merges the predicates forming a single predicate. PathFinder [7] improved on MPF with a re designed filtering engine that was better matched to the ....

M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In Proceedings of the

....This decreases throughput and increases latency. To alleviate these problems, Maeda and Bershad created a library implementation of TCP IP and UDP IP sockets [70] Their library handles the protocol processing for sending and receiving packets and interacts with the network packet filter [139] and network device drivers directly. The library can be linked in with applications that use the networking calls, so each application can do its own protocol processing in its own scheduling domain (i.e. within its own threads) The library only interacts with the UX server to create and destroy ....

M. Yuhara, B. N. Bershad, C Maeda, and J. E. B. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Proceedings of the

....will be insufficient, as the control architecture may wish to apply arbitrary classification policies. Dynamically loadable modules allow the control architecture to exert some influence over the decision process. Classification policies can be expressed in a packet filter language, such as MPF [18], and installed for each VPN at the appropriate (ingress) routers. To avoid the possibility of one control architecture stealing packets belonging to another, and also to ensure that all traffic is accounted to a VPN, a first pass Fig. 13. Multiple control architectures operating ....

M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss, "Efficient packet demultiplexing for multiple endpoints and large messages," in Proc. Winter USENIX Conf., Jan. 1994, pp. 153--165.

....our work, which focuses on provision of QoS guarantees for communication. 25 Handling of incoming traffic: Proper provisioning of QoS guarantees requires appropriate handling of incoming traffic at a receiving host. While a number of recent efforts have focused on efficient packet classification [64 66], attention has also been given to the problem of receive livelock [8] Receive livelock is a phenomenon in which, under network input overload, a host or router is swamped with processing and discarding arriving packets to the extent that the effective throughput of the system falls to zero. ....

M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss, "Efficient packet demultiplexing for multiple endpoints and large messages," in Proc. of Winter USENIX, 1994.

....is often used to discard unwanted packets as early as possible. Past work on packet filters, including the pioneering work on the CMU Stanford Packet Filter [26] a more recent work on BSD Packet Filter (BPF) which uses a register based filter evaluator [23] and the Mach Packet Filter (MPF) [32] which is an extension of the BPF, are related to the work presented in this paper. In the same spirit as packet filtration methods for network monitoring, our approach inserts a filter to intercept messages that arrive from the network. While packet filters are used primarily to gather trace data ....

Masanobu Yuhara, Brian N. Bershad, Chris Maeda, and J. Eliot B. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In Winter USENIX Conference, January 1994. Second Edition. 22

....3 Packet filters are a mechanism for efficiently demultiplexing incoming packets to application endpoints [13] A number of schemes to implement fast and efficient packet filters are available. These include the BSD Packet Filter (BPF) 12] the Mach Packet Filter (MPF) [21], PathFinder [1] demultiplexing based on automatic parsing [11] and the Dynamic Packet Filter (DPF) 3] Most existing demultiplexing strategies are implemented within the OS kernel. However, to optimally reduce ORB endsystem demultiplexing overhead requires a vertically integrated architecture ....

M. Yuhara, B. Bershad, C. Maeda, and E. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Proceedings of the Winter Usenix Conference, January 1994. 7

....hope that any network oriented application program will be able to accomplish acceptable performance. For this reason, packet classification has been targeted by numerous researchers as a potential performance bottleneck that may be amenable to optimization in various forms (see, for example, [1, 3, 7, 9, 10]) Because of the rapid spread and evolution of networked applications in recent years, the quest for high performance packet classification has been paralleled by a demand for the ability to specify and use flexible application specific protocols. The use of application specific protocols, in ....

....classifiers considerably. Historically, packet classifiers have typically been specified using virtual machine instruction sets: this is the case, for example, in the Berkeley Packet Filter (BPF) used in many Unix systems [9] and the Mach Packet Filter (MPF) used in the Mach operating system [10]. This approach has the disadvantage that the specification of a protocol in terms of a low level instruction set can be tedious and potentially error prone, and the hand translated code may or may not exploit all of the optimizations that are possible in a given context. Moreover, the virtual ....

[Article contains additional citation context not shown here]

M. Yuhara, B. N. Bershad, C. Maeda and J. E. B. Moss, "Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages", Proc. Winter 1994 USENIX Conference, Jan. 1994. 12

....capture for developing new protocols without kernel modifications [127] but are increasingly being utilized for realizing protocols such as UDP and TCP as user level libraries. Successive implementations of packet filters have reduced classification overhead significantly, e.g. BPF [114] MPF [186], and PATHFINDER [11] While all of these filter designs perform classification via interpretation, more recently dynamic code generation techniques have been applied to realize very efficient packet filters, as in DPF [59] Packet filters are necessary for packet classification in connectionless ....

....(e.g. the VCI in ATM network adapters) or the necessary information may be carried in link level packet headers [172] in which case the host does not incur any significant packet classification overhead. In the absence of such support, the receiving host must rely upon efficient packet filters [59, 186] to perform the packet classification. Note that this overhead is not incurred at the sending host if the API directly associates outgoing messages with channels. If this is not the case, some classification may be required, but only on a per message basis, since distinct channel handlers shepherd ....

M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss, "Efficient packet demultiplexing for multiple endpoints and large messages," in Proc. USENIX Winter Conference, January 1994. 233

....developed to support efficient demultiplexing in user level implementations of network protocols. They have also been used to enable unobtrusive monitoringon promiscuous mode networks. Existing packet filter implementations are based on interpreters, either stack based [6] or register based [7, 8]. Interpretation is used due to the kernel resident environment of traditional packet filters (for security reasons, most operating systems do not provide convenient ways of compiling and dynamically linking filter code into the OS kernel) Early work on packet filters did not attempt to coalesce ....

....filters into a single composite filter. This approach is fine if few packet filters installed. However, it does not scale when there are hundreds of consumers since the time required to match packets with the filters grows linearly with the number of filters. More recent packet filter research [8] supports a limited form of filter composition. This composition technique is oriented for user level protocols (such as TCP) that contain a common prefix (e.g. the IP header and TCP header fields such as source and destination addresses) followed by variation at the end (i.e. only the TCP port ....

M. Yuhara, B. Bershad, C. Maeda, and E. Moss, "Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages, " in Proceedings of the Winter Usenix Conference, January 1994.

....optimizations A way of improving the speed of the network protocol implementation has been presented by Chris Maeda [13] The idea is that the network code is made into a library that is linked with the application programs. The kernel can parse incoming packets using a packet filter mechanism [22] and send the packets directly to the application program that they are destined to, rather than sending them to UX. See the second example in Fig. 5. The only time an application program needs UX for network traffic would be when a TCP connection is initiated and terminated. UX handles these ....

Masanobu Yuhara, Brian N. Bershad, Chris Maeda, J. Eliot B. Moss, "Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages ", Proceedings of the 1994 Winter USENIX Conference, January 1994.

....packet filters [5] The packet filter used a stack based virtual machine and was general enough to be programmed for new packet types as they became available. Since then, there have been several new packet filters that improve upon BPF, each one better than the last. Mach Packet Filter [6] was an extension to BPF that incorporated caching and the ability to handle fragmented packets. It claimed to be over four times faster than BPF due to its caching of filter results across network ports. Pathfinder [1] implements a declarative language for filter specification. Each predicate is ....

M. Yahara, B. Bershad, C. Maeda, and E. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In Proceedings of the Winter 1994 USENIX Conference. 1994.

....processes, which implement their own protocol code for further processing. This technique was first popularized in the context of microkernel based operating systems (Mach[19] but has since found numerous other applications, such as network monitoring. In most systems (CSPF [15] BPF [14] MPF [21]) a packet filtering engine is a byte code interpreter. The filtering criterion on a network packet is specified as byte code operations involving offset computations and comparisons that amount to a predicate on the packet contents. Recent systems have chosen other filter representations, most ....

....of structure, but do not provide control over layout adequate to be used for specification of arbitrary packet formats. Stub compiler languages such as CORBA [16] or Flick [8] have similar limitations. The idea of a packet filter is certainly not new. Examples include CSPF [15] BPF [14] MPF [21], DPF [9] and PathFinder [2] Of these, only MPF and PathFinder support efficient composition of filters, and they do so by looking for common prefixes in a low level representation rather than by using the layer structure of protocol stacks. Work by Jayaram and Cytron [12] applies context free ....

Masanobu Yuhara, Brian N. Bershad, Chris Maeda, and J. Eliot B. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In Proceedings of the 1994 Winter USENIX Conference, pages 153--165, January 1994.

....of service guarantees. Packet filters are a mechanism for efficiently demultiplexing incoming packets to application endpoints [41] A number of schemes to implement fast and efficient packet filters are available. These include the BSD Packet Filter (BPF) 42] the Mach Packet Filter (MPF) [43], PathFinder [44] demultiplexing based on automatic parsing [45] and the Dynamic Packet Filter (DPF) 40] As mentioned before, most existing demultiplexing strategies are implemented within the OS kernel. However, to optimally reduce ORB endsystem demultiplexing overhead requires a vertically ....

M. Yuhara, B. Bershad, C. Maeda, and E. Moss, "Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages," in Proceedings of the Winter Usenix Conference, January 1994.

....modern BSDderived operating systems. However, many recent innovations in TCP IP implementations are not present. Several research projects have implemented fast demultiplexing packet filters which simplify the process of determining the data destination address on packet reception [DWB 93, YBMM94, EK96, Kay95] Adoption of such an approach should eliminate the associated bcopy and complement other per packet processing improvements. Partridge demonstrated that the per packet CPU cost for UDP processing can be reduced by 26 40 relative to the stock 4.3BSD SunOS 4.1.1 implementation on a ....

Masanobu Yuhara, Brian N. Bershad, Chris Maeda, and J. Eliot B. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In Proc. USENIX Winter 1994 Technical Conference, January 1994.

....microkernel[8] All TCP IP and UDP IP protocol processing are done in the user level library. Figure 4 depicts the mechanism of this approach. The user level library directly accesses a network device driver in the microkernel to send and receive packets. When receiving a packet, the packet filter[10, 19] in the microkernel dispatches incoming network packets to the receiver. Unlike the shared protocol processing server approach, this mechanism does not need context switching to send and receive packets from a network. An operating system server is required only for maintaining sockets. This ....

M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Proceedings of the 1994 Winter USENIX Technical Conference, Jan. 1994. 8

....model, and the fact that I have access to the source code. ffl Mach 3.0 [1] 3 Mach was chosen in part because I am familiar with it, in part because of source access, and in part because it has unique features that are helpful. These features are threads, an efficient packet filtering mechanism [84], and directly accessible network devices. 3 Mach is a microkernel architecture that supports multiple operating system interfaces. In particular, the OS interface that is associated with UNIX has been moved into a user space process known as the UNIX Server (UX) 31] Mach User processes, in ....

....be made at this point concerning the above kernel modifications. Many of the modifications have been made to handle rather special case circumstances (such as the proxy socket) These problems could have been dealt with far more simply using packet filter technology (e.g. Mach Packet Filter [84]) The Low 7 The NFS filter is discussed in detail in section 6.1.5. 4. HIGH LEVEL PROXY DESIGN 57 01: 02: int clientinfd, one, addrlen; 03: struct sockaddrin clientaddr, toaddr; 04: struct inaddr parentaddr; 05: 06: 07: Get the IP address of the MH (client) 08: 09: 10: ....

[Article contains additional citation context not shown here]

M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Proc. 1994 Winter USENIX Conference, pages 153--165. USENIX, January 1994.

....these stack operands. The BSD packet filter (BPF) modernized CSPF with a higher performance register model instruction set. Subsequent research introduced a number of further improvements: the Mach Packet Filter (MPF) extended BPF to efficiently support an arbitrary number of independent filters [24]; PathFinder provided a new virtual machine abstraction based on pattern matching that achieved impressive performance enhancements and was amenable to hardware implementation [2] and DPF enhanced Pathfinder s core model with dynamic code generation (DCG) to exploit runtime knowledge for even ....

Masanobu Yuhara, Brian Bershad, Chris Maeda, and J. Eliot B. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In Proceedings of the 1994 Winter USENIX Technical Conference, pages 153--165, San Francisco, CA, January 1994.

....test to see if this packet filter conflicts with any previously downloaded packet filters. If not, it will be merged with the previous packet filters. The implementation of packet filters, Dynamic Packet Filters (DPF) 14] used in the Aegis and XOK exokernels differs from most other implementations[31, 6, 28, 40] in that it compiles the packet filters to machine code, rather than using an interpreter. This makes installation of packet filters slower than in other systems, but the demultiplexing performance is reported to be on the order of 10 20 times faster than the other systems[14, 13] Fortunately, ....

M. Yahara, B. Bershad, C. Maeda, and E. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In Proceedings of the Winter 1994 USENIX Conference, 1994.

....access pattern information to increase the file system performance that the results are shown in [5] However, their design omits the communication techniques among objects in the system, and, instead, modifies the file system directly. Previous researches in network packet demultiplexing [19, 16, 26] have also used the policy interpretation approach. Applications can program their filter in the filter language, and the system interprets the filter policy and forwards the packet to its destination. The popularity in Packet Filter implementations also shows that the policy interpretation ....

M. Yuhara and B. N. Bershad, `Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages', Proceedings of the 1994 Winter USENIX Conference, San Francisco, California, USA, January 1994, pp. 153-163.

....of service guarantees. Packet filters are a mechanism for efficiently demultiplexing incoming packets to application endpoints [38] A number of schemes to implement fast and efficient packet filters are available. These include the BSD Packet Filter (BPF) 39] the Mach Packet Filter (MPF) [40], PathFinder [41] demultiplexing based on automatic parsing [42] and the Dynamic Packet Filter (DPF) 37] As mentioned before, most existing demultiplexing strategies are implemented within the OS kernel. However, to optimally reduce ORB endsystem demultiplexing overhead requires a vertically ....

M. Yuhara, B. Bershad, C. Maeda, and E. Moss, "Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages, " in Proceedings of the Winter Usenix Conference, January 1994.

....all the way down to the device. Protocols are conceptually replicated, and multiplexing occurs at the lowest layer of the protocol stack. On the receive path, a packet is immediately demultiplexed to the appropriate 11 connection. This demultiplexing is performed by a packet filter or classifier [3, 45, 47, 75]. Given a set of connections, threads, and processors, the assignment or mapping between them can be done in a number of different ways. Choosing a mapping defines the granularity of a connection level parallel implementation. Previous work on connection level parallelism [21, 60, 65, 67] has ....

....can be replicated. For example, TCP connections must be uniquely instantiated. Thus, our CLP implementation requires a packet filter mechanism to demultiplex received packets to the appropriate virtual processor for a TCP (or UDP) connection. Packet filters are becoming more popular (and efficient [3, 75]) in contemporary operating systems since early demultiplexing yields other performance gains. For example, depositing received packets directly into application buffers avoids copying data [15, 41, 70] Our use of a packet filter, to demultiplex to the appropriate virtual processor, simply ....

Yuhara, M., Bershad, B. N., Maeda, C., and Moss, J. E. Efficient packet demultiplexing for multiple endpoints and large messages. In Proceedings of the Winter 1994 USENIX Conference, Jan. 1994.

....service, a library implements a complete TCP IP and UDP IP stack that communicates directly with an in kernel Ethernet device driver. Incoming packets are demultiplexed to application address spaces using the packet filter, a protocol independent packet demultiplexing facility [Mogul et al. 87, Yuhara et al. 94] The library takes raw Ethernet packets from the kernel and does all protocol stack processing before handing the data to the application. Similarly, outgoing data is formatted into TCP IP or UDP IP messages before being sent to the in kernel Ethernet driver. The proper level of distrust is ....

Yuhara, M., Bershad, B. N., Maeda, C., and Moss, J. E. B. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Proceedings of the 1994 Winter USENIX Conference, January 1994.

No context found.

M. Yuhara, B. N. Bershad, C. Maeda, and J. E. B. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Proceedings of the 1994.

No context found.

Masanobu Yuhara, Brain Bershad, Chirs Maeda, and J. Elliot B. Moss. Efficient Packet Demultiplexing for Multiple Endpoints and Large Messages. In Winter 1994.

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