Active Networks Group James P.G. Sterbenz, BBN INTERNET-DRAFT Alden W. Jackson, BBN Category: Experimental Matthew N. Condell, BBN 1 April 2000 HyperActive Networking Status of this Memo This memo defines an Experimental Architecture for the Internet community. This memo does not specify an Internet standard of any kind. Discussion and suggestions for improvement are requested. Distribution of this memo is unlimited. Abstract The cost of processors continues to decrease dramatically, and has resulted in the paradigm shift called Active Networking. The continued decrease in processor cost and ubiquity of smart-everything leads us to propose the next revolution in network technology: HyperActive Networking. This document motivates the technology, proposes a reference architecture, and presents the results of preliminary research spanning the last several hours, consisting of packet formats, performance metrics, security considerations, and potential applications. 1.0 Introduction The cost of processors continues to decrease dramatically. This observation led to the proposition that new network services could be enabled by adding significant processing capabilities to network nodes, and by allowing packets (sometimes called capsules) to contain code to be executed at these nodes [TW96]. Thus, the discipline of active networking was born [CBZS98]. We can now declare that Active Networking is a dramatic success and proven technology, as shown by the quantity of funded research, papers published, and lack of industry interest. The continuing trends in the cost of processing power let us again rethink the [Page 1] Internet Draft HyperActive Networking 1 April 2000 application of computing resource to the network infrastructure. In particular, we can apply processing power to the communications wires themselves, creating smart wires, and propose a field of research: HyperActive Networks (HypeAN). The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL", not used in this document, are to be interpreted as described in [RFC-2119]. 2.0 Architecture A great deal of effort has gone into the construction of the current Active Network reference model [Cal98]. Therefore, we will reuse it to the degree possible, given the time constraints of this architectural effort [1]. Thus, we have WiPEs (wire processing elements) in the wire (copper or fiber), a WireOS, WHEEs (Wire Hyperactive Execution Environments), and WAAHs (Wire Active Applications Hyperactive). There are some key differences from the active node that need to be considered in designing an active wire reference model, shown in Figure 1. Wire ------------------------------------------------------- ( +------+ +------+ +------+ +------+ ) ( | WiPE | | WiPE | | WiPE | | WiPE | ) ( +-----------------------------------------------+ ) ( / *---------- WAAH ----------* / ) ( +- WHEEs ---------------------------------------+ ) ( / *---- WAAH ----* / ) ( +-----------------------------------------------+ ) ( - - - - - - - - - - - - - - - - - - - - - - - - - - ) ( WireOS ) ( - - - - - - - - - - - - - - - - - - - - - - - - - - ) ( +------+ +------+ +------+ +------+ ) ------------------------------------------------------- Figure 1. HypeAN Reference Model WiPEs: In spite of the fact that processing elements continue to greatly decrease in cost, we still expect that the processors embedded in smart wires will be relatively less powerful than those in active nodes. Nonetheless, processing elements will support fully parallel and pipelined operation of the 1 bit payload of each WHEEP packet (discussed in the next section). [Page 2] Internet Draft HyperActive Networking 1 April 2000 Furthermore, the WiPEs need to support line rate filtering of WHEEP packets, including the ability to filter on arbitrary, complex filter specifications of the payload. Packet filters need to be able to snarf, copy, or turn WHEEP packets around to the opposite direction of the wire [2]. WireOS: Since the WiPEs are relatively less powerful than conventional (hypo-) active node processors, the WireOS will be linearly distributed along the entire length of the wire. This is necessary to support a fully functional, multi-threaded, multi- tasking, virtual-memory, window-GUI, multimedia operating system [3]. WHEE: WHEEs can similarly be distributed along the length of the wire. P-WHEEs (permanent WHEEs as assigned by the grand exalted highness of the HYIANA) [RB00] will be distributed over the entire length of the wire, as in the case of the WireOS. Other WHEEs can be instantiated and terminated as necessary, and thus flow along the wire with the WAAHs which they execute. WAAH: WAAHs similarly are created and terminated as necessary, and thus flow along the wire in WAAH-windows along with the code they execute. The mobility of WHEEs and WAAHs is a key difference from conventional (hypo-)active networks; this feature has implications to potential PhD students that are staggering, indeed. 3.0 Packet Formats The goal is to provide an efficient mechanism whereby WAAHs and WHEEs can perform computation on the payload while traversing the communication medium. The described solution includes defining an encapsulation protocol to carry the single bit/packet payload. WHEEP (Wire HyperActive EE Protocol) payloads are one-bit each, to allow maximum flexibility in the processing by WiPEs, and to eliminate arguments over the optimal payload sizes. Bandwidth has become so cheap, that header overhead of 289:1 is not significant. Active nodes will fragment/reassemble conventional ANEP datagrams to WHEEP packets. The WHEEP packet payload consists of a ha-bit, ra-bit, or qu- bit. The bit is encapsulated in the WHEEP, an ANEP frame (for compatibility with the existing Active Network architecture), a transport protocol frame and any appropriate lower layer framing. [Page 3] Internet Draft HyperActive Networking 1 April 2000 +-----+----+-----------+------+-------+---------+ | MAC | IP | UDP / TCP | ANEP | WHEEP | Payload | +-----+----+-----------+------+-------+---------+ +-------+-----+----+-----------+------+-------+---------+ | SONET | ATM | IP | UDP / TCP | ANEP | WHEEP | Payload | +-------+-----+----+-----------+------+-------+---------+ +-------+-----+----+----+-----------+------+-------+---------+ | SONET | ATM | FR | IP | UDP / TCP | ANEP | WHEEP | Payload | +-------+-----+----+----+-----------+------+-------+---------+ The ANEP Type ID for WHEEP is 1010111110101101 (0xafad). The format of the WHEEP header is: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Version | Type ID | Context ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | | | | | Serial Number | | | | | | | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P| +-+ Version: A 1 octet code in Intel(TM) byte order indicating the WHEEP version number. Type ID: A 1 octet code in Intel(TM) byte order indicating the type of bit in the payload with the following values: 0 ha-bit 1 ra-bit 2 qu-bit 3-255 Reserved to HYIANA for future use. Context ID: A 2 octet unique number, in Intel(TM) byte order, on the bit originator to identify the application that generated this bit. Multiple bit generating applications could be running on the same source, unless it has an operating system from Redmond, WA. [Page 4] Internet Draft HyperActive Networking 1 April 2000 Serial Number: A 64 octet unique sequence number, in Intel(TM) byte order, assigned by the source to identify this bit. This field is able to enumerate the proposed number of hydrogen atoms in the universe, give or take a galaxy or two, or the IPv6 address space with room to spare [4]. P: Payload, 1 bit, in Intel(TM) byte order. 4.0 Performance Metrics While all the usual conventional and active networking performance measures apply to the HypeAN environment, there are two new key metrics that need to be considered in performance studies: P/m: This is the number of processing elements per linear meter of wire, and indicates the density of processing capability. BW-x-d-x-P: This is the bandwidth-x-delay-x-processor product, and refers to the product of the conventional bandwidth-x-delay product (in bits) and the total processing capability of the link (in floating point operations per second). Thus the standard dimension of this unit is yotta-bitflops [5]. 5.0 Security Considerations Security exposures are no worse than the product of general network security, active networks security, and (in the case of wireless HypeAN) mobile and wireless security issues. Securing the WHEEP payload will require single-bit cryptography to authenticate the sender of the ha-bit. A WAAH can use one-bit encryption (OBE) to hide its ha-bit from others. Several hashing and encryption mechanisms have been developed to provide single-bit security. 5.1 Hash Techniques Zeroing (HaZe): Hash the ha-bit x using the function f(x)->0. Rumors claim that a collision has been found in HaZe, however the discoverers still seem to be lost. [Page 5] Internet Draft HyperActive Networking 1 April 2000 5.1 Hash Techniques Bit-flip: XOR the ha-bit with one (1). Very secure when an attacker does not know the algorithm that is being used. Bit-leave: The bit-leave technique (BLT) is the equivalent of two rounds of the Bit-flip algorithm. One-time pad: A random bit (ra-bit) is handed off (hop'ed) prior to sending a WHEEP packet. The ra-bit that hop'ed is XORed with the ha-bit to encrypt it. This is equivalent to performing a random number of rounds of the bit-flip algorithm. In the near future, quantum computing will provide better techniques for single-bit cryptography. The ha-bit will be encrypted as a single qu-bit. A preliminary implementation of the ha-qu transformation has been completed. The one-bit encrypting WAAH that does qu-bit network one-bit encryption (OBE WAAH QNOBE) will be designed once the ha-qu transformation has been perfected. 6.0 Potential Applications One of the most promising applications of HypeAN results in a dramatic reduction in network latency. By overlapping protocol processing with the transit of bits through the wire, the latency of processing at a node (router or switch) can be dramatically reduced. In fact, this eliminates one of the major concerns of active networking, which is the additional latency at node to do active processing [6]. A number of useful network services can be envisioned, including congestion control and traffic shaping in the *middle* of a link, or multicast by replication on additional wavelengths in a fiber. The provision of QoS and reliable multicast are problems that are clearly solved by HypeAN. Previous researchers have considered how to exploit the storage characteristics in high bandwidth-x-delay product links. The expectation of the Interplanetary Internet provides us with the significant challenge of extremely high bandwidth-x-delay products. For example, a Web browser on Mars faces irritating delays for Earth- based content which is not locally cached. HypeAN allows content to be arbitrarily cached in the wireless link, with WiPE packet filters dynamically creating storage loops of content along the communication channel, moving the loop boundary based on application demand. Note that wireless HypeAN requires smart aether, which is beyond the scope of this document. [Page 6] Internet Draft HyperActive Networking 1 April 2000 Security is a particular concern in mobile wireless networks, and it is important for these networks to be highly adaptive in routing and policy. HypeAN provides new opportunities in this area, for example for the channel itself to detect eavesdropping, encrypt the packet payloads, and then decrypt once the channel has left the malicious area. Numerous other military and commercial applications certainly exist, which provide payoff to funding agencies far in excess of the establishment of a large HypeAN research program. 7.0 HYIANA Considerations The assignment of the P-WHEE status to a WHEE is performed by the HYIANA (or HYBOCC). The WHEEP header Version and Type ID are assigned by the HYIANA. New values are to be assigned with the consensus of the DARPA Active Networking Community or by the fiat of the program manager [WDM00]. 8.0 Notice on Intellectual Property Intellictual Property concerns are not addressed in this document. [Page 7] Internet Draft HyperActive Networking 1 April 2000 9.0 References [Cal98] K. Calvert, ed., "Architectural Framework for Active Networks", AN draft, AN Architecture Working Group, 1998. [CBZS98] K. Calvert, S. Bhattacharjee, E. Zegura, and J. Sterbenz., "Directions in active networks", IEEE Communications Magazine, 36(10), October 1998. [RB00] Robert Braden. [RFC-2119] Bradner, S., "Key words for use in RFC's to Indicate Requirement Levels", Internet Request For Comments No. 2119, March 1977. [RFC-2434] T. Narten and H. Alvestrand, "Guidelines for Writing an IANA Considerations Section in RFCs", Internet Request For Comments No. 2434, October 1998. [RFC-2460] S. Deering and R. Hinden, "Internet Protocol, Version 6 (IPv6) Specification," Internet Request for Comments No. 2460, December 1998. [TW96] D. L. Tennenhouse and D. J. Wetherall, "Towards an Active Network Architecture", ACM Computer Communication Review, April 1996. [WDM00] W. Douglas Maughan. 10.0 Notes [1] We realized Friday afternoon 31 March that the deadline for this document was the next day. [2] Bidirectional fiber strands are beyond the scope of this document. [3] The potential requirement of a Win-32 API is beyond the scope of our comprehension. [4] Actually, the proper calculation, which we didn't have time to do, is to calculate the number of payload bits transmittable over a link for the remaining life of the universe. [5] Not to be confused with 'lotta-bitflips' which is the result of some of the payload encryption schemes described in Section 5. [Page 8] Internet Draft HyperActive Networking 1 April 2000 [6] The other major objection being "its different than we do it now in the Internet". 11.0 Authors' Addresses Matthew Condell Phone: +1 617 873 6203 BBN Technologies Email: mcondell@bbn.com 10 Moulton Street Cambridge, MA 02138 USA Alden Jackson Phone: +1 617 873 2126 BBN Technologies Email: awjacks@bbn.com 10 Moulton Street Cambridge, MA 02138 USA James Sterbenz Phone: +1 508 944 3067 BBN Technologies Email: jpgs@sterbenz.org 10 Moulton Street Cambridge, MA 02138 USA [Page 9]