|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||
|
||
|
||
|
||
|
 |
 T1 Inverse Multiplexing: Getting Started on the Road to ATM Contents
Enterprise network users don't need to be told that bandwidth pressures are one of the main factors in the continuing evolution of their networks. Various applications are straining the typical T1 WAN connections between networked sites, including:
Initially, corporations faced with this bottleneck adopted load sharing techniques to interconnect networks. Gaining popularity now, however, is inverse multiplexing, which provides a single high-speed channel using multiple T1s. One impetus for the popularity of inverse multiplexing is the way it prepares users for the adoption of ATM. Inverse multiplexing shares with ATM many important similarities which impact the design of network architectures. This allows users of inverse multiplexing today to enjoy some of the benefits of ATM, and ensures an easy migration to ATM services when they are available. Major Distinctions Between Inverse Multiplexing and Load SharingTo understand why inverse multiplexing has gained popularity, let's look at the basic characteristics of this technique versus load sharing. (See Figure 1.) Inverse multiplexing is a process whereby multiple T1 lines are combined to create a single logical data channel that is the aggregate of the T1 bandwidths, minus a small amount used for inverse multiplexing overhead.The function of the inverse multiplexer is to divide a high-speed serial data stream from a router or other device into partial data streams of approximately 1.5 Mbps each, transmit these partial streams across separate T1 lines, and recombine the partial streams into the exact original stream at the far end. From the point of view of the routers or other devices connected to the inverse multiplexers, they are communicating via a single high-speed WAN channel at some multiple of the T1 rate. In other words, they have the bandwidth of a fractional T3 data service using readily-available T1 services. Figure 1. T1 inverse multiplexing and load sharing are similar in the way that
they employ multiple T1s for WAN connections, but differ in the way they employ the T1s.
The inverse multiplexer combines the T1s into a single clear data channel, while the
load-sharing router treats the T1s as separate paths across the WAN. In a diagram, load sharing looks very similar to inverse multiplexing. Multiple T1 lines are used to provide increased bandwidth linking two routers. However, there are key differences between inverse muxing and load sharing, which depend in part on the way load sharing is implemented by the router. Many routers, including those from Cisco and Bay Networks (Wellfleet), employ"route caching" to determine how packets will be routed across the available T1s.By this method,the router assigns each particular session a particular T1 port when the session is initiated. This makes it easy for the router to direct packets to T1 ports, but it limits each application to no more than 1.5 Mbps of WAN bandwidth. Thus, while inverse multiplexing of four T1s and load sharing using four T1s offer roughly equivalent total bandwidth between the routers, the bandwidth available to a particular application is four times greater using inverse multiplexing. Another advantage of inverse multiplexers over load-sharing routers using route caching involves the ability of applications to recover from T1 failures. Inverse multiplexers such as Larscom's Mega-T and Orion 4000 support automatic rate recovery and restoral. In other words, they have the ability to maintain the data channel in the event of T1 failures, simply reducing the channel's bandwidth to that supported by the remaining T1s. At worst, applications will experience a slight"hiccup" and perhaps slower performance for the duration of the T1 failure. With load-sharing routers employing route caching, a T1 failure may cause the application session to time out and abort, generally requiring operator intervention to restart it. Instead of route caching, some routers employ frame-by-frame load sharing. The transmitting router monitors each T1 port and routes each packet individually according to the availability of the T1s. The receiving router gathers packets from all T1 ports and routes them to the applications, in some cases having to reorder packets which were received out of order due to variable delays among the T1 links between the two routers. While this method does overcome the bandwidth-per-application limitation of route caching, it has its own disadvantages: The amount of processing performed at each end lowers throughput and raises latency (end-to-end delay). Latency can be especially problematic because many of the high-bandwidth applications are latency-sensitive, and latency effects on a data channel are inversely proportional to the data rate. One more advantage of inverse multiplexing---and not an insignificant one---is that it uses less equipment. Instead of requiring multiple serial ports as load sharing does, an inverse mux employs a single router serial port. In many cases this frees up valuable card slots on the router for other uses. Also, a single inverse mux replaces multiple T1 DSU/CSUs. From a management standpoint, this simplicity can become quite important. For example, if SNMP is being used to manage the devices, only one IP address is required for the inverse mux, whereas each T1DSU/CSU in a load-sharing implementation will require its own IP address. Enjoying Some of ATM's Benefits Before ATMUsers contemplating a choice between load sharing and inverse multiplexing are looking at these issues we've been discussing, along with issues of price, redundancy, legacy equipment, etc. One issue that is frequently overlooked, however, is the relationship between inverse multiplexing and ATM. In a nutshell, ATM is a cell-based telecommunications technology which can implemented at a variety of rates over a variety of media. Its chief benefits include a high degree of scalability (or granularity) and the ability to seamlessly link LANs and WANs in individual enterprise networks. ATM's benefits have received a great deal of discussion in the industry press, so we will not go into them here other than to provide these basic definitions:
These benefits of scalability and seamlessness are, to a lesser degree, among the benefits of inverse multiplexing. With both ATM and inverse multiplexing, the WAN connection is a single data pipe which can be divided among various applications as the user desires. (See Figure 2.)This is a straightforward statement of a simple concept, but its implications are highly significant to today's network designers and managers. The similarity of topology between ATM and inverse multiplexing -- individual sites linked by "clear-channel" broadband data pipes -- means that applications can be looked at and implemented in much the same way when either technology is used. Figure 2. T1 inverse multiplexing and ATM are alike in providing a clear
broadband channel across the WAN.
Load sharing, especially where route caching is employed, uses a different topology than ATM and inverse multiplexing. The multiple T1 lines do not provide a single clear data channel, but rather a set of parallel T1 channels. From the point of view of high-speed applications, this is a major difference, and it impacts the way network designers look at and implement these applications. Figure 3 shows the basic difference between an ATM or inverse multiplexing network and a load sharing network in the way a particular application might be implemented. The application involves a graphical data base which is shared and modified by several users at two different network sites, with frequent large file transfers. Performance is noticeably degraded at bandwidths of less than 3 Mbps between workstations and the host. - Figure 3. T1 inverse multiplexing, by offering multi-megabit low-latency WAN
connection between applications, supports implementations not supported by load-sharing
routers. Here the high bandwidth needed for workstation/host operations necessitates
duplication of hosts in the load-sharing environment.
In an ATM or inverse multiplexing network, this application can be supported easily by a host atone site or the other. Because applications have access to bandwidths of 3 Mbps and greater,there is little or no performance difference between workstations on the local and remote LANs. The situation is not so simple with load sharing, however. Because the WAN connection is limited to 1.5 Mbps (one T1) per application if route caching is employed, or is subject to significant delay if packets are routed over different T1s, a workstation communicating with a host on a remote LAN will find that performance is significantly degraded relative to the performance at a workstation on the same LAN as the host. Given this situation, the network manager's options include:
Some network managers will have the luxury of ignoring the problem until ATM comes along.But for those with pressing bandwidth demands today, the solution ATM offers in the future is not much help. For them, implementing inverse multiplexing can give them some of ATM's benefits before ATM is on the scene -- and make the changes to ATM easy when it does occur. Migrating from Inverse Multiplexing to ATMWith inverse multiplexing, migration to ATM is easy because the "clear high-speed data pipe"topology means that applications are implemented in the same way. As Figure 2 suggests,migrating from inverse multiplexing to ATM can be as simple as replacing a pair of inverse multiplexers with a pair of ATM DSUs when the carrier makes ATM service available. There is less need to rethink the way applications are implemented because the nature of the network ha snot changed. On the other hand, replacing T1 load sharing with an ATM DSU represents a significant change to the nature of the network that may necessitate a costly and time-consuming rearrangement of applications across the network. It is also quite likely that inverse multiplexing will play a role in the access options carriers offer to their ATM services. Figure 4 shows a likely scenario, where the carrier allows access to its ATM hub at fractional T3 rates using inverse-multiplexed T1s. In cases like this, the user will not even need to install an ATM DSU to migrate to ATM; the only thing necessary is to sign up for the carrier's ATM service. Figure 4. T1 inverse multiplexing is likely to be offered as an ATM access
option by some carriers, meaning that no equipment changes will be necessary to migrate
from current T1 inverse multiplexing to ATM. Similarly, the user may install an ATM hub at a major network site, and use inverse multiplexing to extend the network to a remote site where ATM service is either unavailable or uneconomical.Figure 5 illustrates such an application, with Larscom's shelf-based Orion 4000 Broadband Access multiplexer serving as the ATM hub at the main site. One module installed in the Orion4000, an IMUX T1 inverse multiplexing module, provides a multi-megabit link to a smaller site employing Larscom's stand-alone Mega-T T1 Inverse Multiplexer. Figure 5. T1 inverse multiplexing can be used to extend an ATM network. Here a
Larscom Orion 4000 serves as an ATM hub and main networking site. An IMUX T1 inverse
multiplexing module in the Orion 4000 supports network extension to a smaller non-ATM site
via inverse-multiplexed T1s and a Larscom Mega-T stand-alone T1 inverse multiplexer. Despite all of the talk and ink generated by ATM, and all of the rosy promises it embodies, no one really knows how and when ATM will emerge as a major networking reality. For those trying to balance today's needs with tomorrow's promises, inverse multiplexing offers both immediate benefits and smooth migration. It allows bandwidth-pressured but wary network designers and managers a safe and secure way to begin the journey to ATM. Larscom Incorporated Tel: 408.988.6600 |
 |
|
|
|
|
|
|
|
|
