TotalTransport Technical White Paper

Predictable Application Response at Any Network Distance with TotalTransport™ Technical White Paper Predicable Application Response At Any Network Distance with TotalTransportTM Page 2 Executive Summary With application workflows traveling over standard IP networks across states and continents, remote workers need fast, predictable response times. That’s what TotalTransport1 technology delivers. With an Internap® Flow Control Xcelerator solution placed at each end of a network circuit, workers get LAN-like throughput and reliability across the wide-area circuit. That kind of performance is now a necessity for collaborating beyond the office LAN. Fast throughput is required for manufacturers doing collaborative design and engineering, media-related companies exchanging digital media assets, and every company that transfers large files over long distances as part of a collaborative workflow. TotalTransport is application-and protocol-independent: it works for any application running over an IP network that utilizes standard TCP-based flow control. TotalTransport works for all business application traffic, all the time. A key benefit of the technology is that it will safely and effectively fill the pipe provisioned between any set of offices. TotalTransport is needed because of a phenomenon associated with traditional TCP/IP-based networks: as distance and bandwidth increase, the effective throughput of the network decreases. This throughput degradation, which seems counter-intuitive, is an unintended consequence of the simple, but elegant algorithms that govern IP networks and have enabled the Internet’s extraordinary growth. TotalTransport provides a standards-based improvement to the original IP design, enabling business application traffic to fully utilize the network resources at their disposal. TotalTransport fills the pipe for all IP traffic all the time. 1 TotalTransport Technology is a patent-pending technology developed by Orbital Data. Predicable Application Response At Any Network Distance with TotalTransportTM Page 3 Understanding Network Application Response Time IP networks, including the Internet, use the TCP protocol for most application communication. TCP is short for Transmission Control Protocol. The IP protocol specifies how packets are routed, while TCP handles flow control, congestion control, retransmissions, and in-order delivery of data. The IP protocol actually is shorthand for a whole family of protocols, usually referred to as the IP protocol stack. The IP stack consists of layers. Each layer corresponds to a different aspect of communication. The point of designing IP this way is that it is flexible and protocols built on top of it do not have to be monolithic. For example, the FTP protocol running on TCP/IP is only concerned with the transmission of files and does not need a full network protocol implementation. The IP networking protocol was developed in the early 1970s by pioneering engineers Bob Kahn and Vinton Cerf. By 1983, ARPAnet, the forerunner of today’s Internet, adopted TCP/IP since experts realized that the adoption of a single networking protocol would be an important step toward maintaining order within the growing community that was to become the Internet. TCP/IP provided a technological bridge for small networks to connect to the Internet much more easily than before. TCP is brilliant engineering, but it must be viewed in the context in which it was conceived. The Internet was originally designed to facilitate relatively low-bandwidth communication and asynchronous data sharing between government and education facilities. When TCP was first tested and implemented, the dominant application was email with its very small payloads, and WAN circuits were measured in the hundreds of kilobits. With TCP as it was conceived, and as it is implemented today, flow control is confined to the endpoints of a network. But it’s in the interior of the network where most of the problems occur that induce latency, packet loss, and unpredictability. With conventional TCP, only the endpoints communicate to each other. As the distance between the endpoints increases, so does the amount of time it takes for the network to respond to problems occurring anywhere between. Furthermore, using conventional TCP, only the state of the individual connection between the sender and receiver is factored into the calibration of the data flow between them. The endpoints are not aware of the other conditions around them, or within the network, until something “breaks” (such as packet loss or round trip timeouts). To someone gauging network performance from the outside, this usually is seen as a fairly gradual increase in latency and diminution in response time. As seen in Figure 1 as the distance between any two endpoints (Client A and Server A) is increased the round trip response time of any packet is increased thereby reducing the sensitivity of the system. Furthermore if the problem or congestion in the network is actually being caused by Client B and Server B then the Client A and Server A endpoints are unaware and will be adversely affected. Conventional IP networks can be too slow for today’s business-critical collaborative workflows that occur across remote offices over public or private IP networks. These problems with conventional TCP/IP and legacy flow control become apparent when data flowing over a WAN encounters bandwidth and latency transitions, for example, when bridging to a LAN (Local Area Network). Congestion occurs, resulting in packet discards that cause throughput and transaction times to vary unpredictably. Moreover, latency between the endpoints and the congestion points becomes a problem. Latency makes a control loop particularly hard to manage, because response data is not real-time. A good way to think of this is to picture adjusting the hot and cold water in your shower. Because of the delay in the pipe’s reaction to your adjustments, the temperature must be adjusted very slowly to avoid getting too hot or too cold. The endpoint-only transport implementation of TCP causes the same sluggish response. Clearly, endpoint-only control is not good enough to speed flow control. What’s required is adaptive behavior at the point of congestion, enabling networks to establish a feedback loop without undue delay due to latency. Network technologies, such as SNA (Systems Network Architecture) and Fibre Channel, solve this problem, generally by putting transport control at every hop in the network. But these solutions are problematic because it results in networks that may not be Predicable Application Response At Any Network Distance with TotalTransportTM Page 4 Figure 1: The legacy flow control provided by a conventional IP network causes slow and unpredictable application response time. TCP/IP compliant and are difficult and expensive to deploy and maintain. Collaborative Workflows on the Internet Business application use of WANs has evolved over time, as has network bandwidth (see Figure 2). It surely comes as no surprise to note that as time goes by businesses are making more use of WANs, and network bandwidth is increasing. The original use of wide-area networks was for email and asynchronous communications. Over the succeeding years, more complex applications evolved, such as Customer Relationship Management (CRM) applications and interactive Web applications. Many of these applications involve transferring larger files. These applications required far better throughput to perform well than earlier “svelte” applications, and in response, several WAN optimization techniques evolved: most notably, caching and compression. These techniques attempted to address bandwidth limitations by either not re-transmitting payloads (caching) or compressing them so that they require less resources. Caching and compression are very effective tools for solving some problems associated with bandwidth limitations for asynchronous communications and transactional applications. But they are not suitable for all network traffic, and they do not work all the time. While it’s obvious that business usage of WANS, and network bandwidth is increasing, what is less obvious is that today’s collaborative workflow process has enormous consequences, often involving the transport of large digital payloads over multi-megabit circuits. Examples include: • Collaborative engineering with Product Lifecycle Management (PLM) products Predicable Application Response At Any Network Distance with TotalTransportTM Page 5 Figure 2: Increasing network bandwidth over time. • • Digital media workflows involving Digital Asset Management (DAM) products Remote visualization of three-dimensional data in medical workflows and geo-physical analysis. These applications are increasingly in need of a delivery infrastructure that will enable them to fully utilize the available bandwidth between offices. With the emergence of metro-area Ethernets and MultiProtocol Label Switching (MPLS), wide area IP networking is now evolving into “transparent LAN bridging.” Here, the goal is to merge LANs and WANs into one very high bandwidth network “cloud,” where all applications behave as if they were on a LAN and distance is not a factor. One application that stands to benefit from this trend is IP storage, where the widearea distributed file-systems behave as if they were LANs interconnected over IP. Tomorrow’s transparent LAN-bridging world requires underlying IP networks that operate at gigabit speeds and beyond. These networks cannot afford degradation of response time. In each of these kinds of applications (and many others that may or may not involve transmitting large-sized files) neither caching nor compression may be an option, and bandwidth limitations are often not the underlying cause of slow, unpredictable throughput often – but certainly not always – when large-sized files are involved. The real culprit is the ineffective use of the bandwidth at a distance. In other words, the pipes may be there, but they are not being utilized. There is a great need to effectively and safely utilize the bandwidth that is already in the pipe. Predicable Application Response At Any Network Distance with TotalTransportTM Page 6 The Problem Refined With traditional endpoint-controlled TCP networks, as distance and bandwidth increase, the effective throughput of the network decreases. This throughput degradation, which may seem counter-intuitive, is an unintended consequence of the simple, but elegant algorithms that govern TCP operations. In part, these algorithms have enabled the Internet’s extraordinary growth. To send a transmission over TCP, an endpoint controller doesn’t have to know anything about the internals of a network or the nature of the transmission recipient. Absent relatively gross error conditions such as packet loss or roundtrip timeouts, the endpoint controller just keeps on transmitting with no knowledge of throughput degradation on the network. This absence of knowledge of network internals, and lack of quality of transmission communication with message recipients, is probably a necessary requirement for a flexible, public IP network. After all, you don’t want to have to know about the equipment your public Internet transmission will be traversing, or the hardware or software at the other end. But this very flexibility of architecture leads to bandwidth degradation as transmissions hit internal bottlenecks without adequate feedback. To drill down on this point a bit further, in a conventional IP network, an interior node handles a bottleneck by dropping packets. The network then relies on the end-points to detect the loss, recover from it, and adapt network conditions to prevent further loss. The scheme works fine within the confines of a LAN. But over the longer reaches of a WAN, edge intelligence is not enough – in part because it takes so long for the edge nodes to become “aware” that a problem has occurred deep inside the network. The TotalTransport Solution TotalTransport is implemented within two or more FCX units and uses advanced flow control, retransmission, and congestion control algorithms in a point-to-point or many-tomany mesh as shown in Figure 3. The idea is to enhance standard IP networks with an updated, highly sophisticated implementation of flow control. TotalTransport, the technology within the FCX units, is a TCP/IP Layer 4 implementation that can accelerate all IPbased traffic, including traffic using FTP, HTTP, SMTP, and NFS/CIFS, and other protocols. The TotalTransport implementation consists of highly sophisticated transport control algorithms that accelerate all the traffic all the time in a way not possible with caching and compression. Rather than just putting the transport intelligence only at the edges of the network, the flow control supported by TotalTransport places it at the network transition points where congestion most often occurs. Examples of these critical congestion points include: • Bandwidth transitions, such as a megabit pipe to a remote office coming off a gigabit LAN, or at the connection from a WAN to a LAN Latency transitions at the end of a long link, for example when public TCP/IP is being used to transmit digital assets to a remote site thousands of miles away Data links subject to media losses, such as at a wireless network hub because wireless transmissions are subject to greater packet loss • • TotalTransport divides the end-to-end control loop into sections that are managed independently. WAN flow control is optimized for long-distance transport using powerful algorithms developed following an extensive research and development effort. The algorithms employed by TotalTransport to enhance flow control are an extension to the classic TCP/IP model. So the good news is that TotalTransport, unlike other performance enhancements, is fully TCP/IP compatible. This means that TotalTransport solves one of the biggest problems with other performance-enhancing approaches: network incompatibility. Other approaches convert TCP/IP packets into proprietary formats. Doing so may enhance flow control performance, but it does so at the cost of making the packets unintelligible to firewalls, intrusion detection systems, load balancers, network monitors, and other network equipment. Predicable Application Response At Any Network Distance with TotalTransportTM Page 7 Figure 3: Placing FCX units at strategic network locations results in highly responsive and adaptive flow control mesh. Since TotalTransport is a Layer 4 TCP/IP implementation, packets are fully TCP/IP compatible from end to end, preserving your investment in equipment and operations, and making it easier and less expensive to maintain your network – now, and in the future. TotalTransport addresses the need to deliver fast, predictable response time as part of a collaborative workflow process, often involving the transport of large digital payloads over multi-megabit circuits. As discussed earlier in this White Paper, these large digital file transfers are becoming an increasingly significant part or modern collaborative business practice. TotalTransport enables these collaborative business workflow applications to more effectively use the available bandwidth and provide reliable, fast response times at each step of the workflow process. Today and tomorrow’s real-time multimedia applications will increasingly consume available bandwidth. TotalTransport can help you provision your bandwidth to insure that standard TCP-based applications don’t “step” on voice and video streams, thereby causing the annoying stutter so often experienced today. Inside TotalTransport As noted earlier in this White Paper, a deployment of the TotalTransport technology requires the use of at least two FCX units. The units should be installed at potential bottlenecks, for example LAN/WAN connections. It’s a great benefit that the TotalTransport technology can be deployed incrementally. You can start with two units at the most critical transition points, then add additional units later on as the benefits of TotalTransport become apparent, and as your network traffic increases as seen in Figure 4. With TotalTransport, IP packets pass through an FCX unit. Each FCX unit has three functional components: Predicable Application Response At Any Network Distance with TotalTransportTM Page 8 • • • A receiver, which accepts the packets passed to it on the LAN or WAN A sender, which sends packets on through the network to another FCX unit on the WAN or to a LAN destination The deep-packet inspection and policy engine Figure 4: TotalTransport can be used in a point-to-point or many-to-many mesh depending on the deployment. The deep-packet inspection and policy engine is the heart of the FCX TotalTransport unit. Packets received by the FCX unit receiver are passed through the application-aware deep-packet inspection engine to the sender, where they are sent on. The deep-packet engine provides the intelligence needed to decide which packets get sent, and when. The algorithms make a policy decision about which packets get sent first. This policy is derived from two factors: • • Management policy, driven by the deep-packet inspection conducted by each FCX unit individually. The state of the entire network, based on information collected by all FCX units on the network. This breadth of information is what makes TotalTransport so effective at the important task of allocating network bandwidth carefully. When the packets arrive, the engine not only derives substantial information about the specific connection that the packet traversed, but also has available the broader view of network connections, as gathered by other FCX units. Thus, the information exchanged between FCX units is much richer than what is utilized at the edges of a conventional IP network. An active feedback mechanism, combined with informed management decisions about which packets are sent first, result in greatly improved network performance (your WAN will feel like a LAN!). TotalTransport provides a rich communications link between FCX units, which are best deployed strategically at network transition points. As a Layer 4 TCP/IP implementation, TotalTransport provides the required services in a way that is fully and completed TCP/IP standards compliant. The FCX units make intelligent, WAN-optimized transport decisions, resulting in more efficient allocation and full utilization of bandwidth. When that happens, throughput approaches Predicable Application Response At Any Network Distance with TotalTransportTM Page 9 the limit of the bandwidth available in the pipe without the corresponding degradation in response-time predictability that occurs with standard IP networking. This results in the best possible network performance and application response time. Network utilization rates go up, and the pipes are safely and effectively filled. TotalTransport is application and protocol independent. It works for any application running over an IP network that utilizes standard TCP-based flow control. TotalTransport works for all business application traffic, all the time. It is can be used in concert with or instead of other optimization strategies, such as caching and compression. TotalTransport technology will safely and effectively fill the pipe provisioned between any set of offices, no matter whether the public Internet is used or whether a private TCP/IP infrastructure is deployed at one or both ends. Conclusion IP networks provide inadequate intelligence about internal bottlenecks, making it hard to effectively utilize available bandwidth resulting in slow application response time. Optimization techniques such as caching, compression, and application protocol acceleration handle only some of the network traffic some of the time and/or create IP network compatibility issues. The TotalTransport technology implemented in FCX units is fully TCP/IP compatible. TotalTransport provides intelligence about bottlenecks deep in the heart of the IP network, and helps speed all the traffic, all the time. Organizations that make effective use of TotalTransport FCX units can increase the efficiency of their networks with relatively little effort, and help fill their pipes to capacity safely and effectively. TotalTransport provides predictable and fast application response time. For more information contact: Internap Network Services 250 Williams Street Atlanta, GA 30303 Tel: 404.302.9700 or 877.THE.PNAP Fax: 404.475.0520 © 2004 Internap Network Services Corporation with permission from Orbital Data Corporation. All rights reserved. Internap is a registered trademark of Internap. Total Transport is a trademark of Orbital Data. All other trademarks and brands are the property of their respective owners.