Efficiency isn't just a pipe dream
New products help net managers squeeze more bandwidth out of systems
By Erick Von Schweber for PC Week Online
Network managers and ISPs can't easily expand the pipes that connect their remote users and interconnect their LANs, but some new products can help get more effective bandwidth out of the pipes they already have.
Given the sizable expense of today's communications channels, these technologies promise a rapid payback, making them worthy of attention. Network managers looking to improve network performance should consider their needs and choose a technology and vendor(s) accordingly.
Remote users running an IPX network with Version 4.5 or later of a Shiva Corp. RAS (remote access server) or switch already have the PowerBurst server and need only activate it and install client software. Others should look to the Point B Remote Net-Accelerator from Traveling Software Inc., which provides remote access to IP and IPX networks over dial-up lines, WANs and VPNs (virtual private networks).
Managers who must accelerate LAN-to-WAN connections as well as Internet service provider Web server farm access and client/server applications should look to RUN Inc.'s runTCP, a zero administration solution.
Remote Access Caching, employed in Point B Remote Net-Accelerator, in Shiva's PowerBurst-enabled LANRover RASes and switches, and in TurboGold from Stampede Technologies Inc., helps reduce the volume of data actually transmitted between a remote client and a network server, without affecting performance.
All of these implementations require client software to be installed on all remote machines, and software, hardware or both to be installed on the corporate LAN. The requirement for special client software makes this a poor choice for an ISP. However, Point B is compatible with VPNs, so it could be packaged with RAS outsourcing by an ISP and can scale by adding Point B servers to the corporate LAN.
Enhanced Transport Control technology, used in runTCP, seeks to quickly push actual bandwidth closer to its theoretical maximum by keeping IP traffic moving smoothly while minimizing retransmissions. Enhanced Transport Control works best where fast and slow networks meet.
Remote Access Caching and Enhanced Transport Control complement another recently introduced technology, Bandwidth Partitioning, offered by Aponet Inc.'s Bandwidth Manager, Packeteer Inc.'s PacketShaper and Xedia Corp.'s Intelligent Queuing. Bandwidth Partitioning divides available bandwidth and assigns its use, thereby reducing the possibility of oversaturating the pipe.
Organizations with severe congestion problems would do well to evaluate Remote Access Caching, Enhanced Transport Control and Bandwidth Partitioning together, because the technologies could complement one another and effectively unclog pipes.
Remote access caching
Remote users whose computer and home office network are equipped with a remote access cache might see little performance improvement the first time they connect to the home office network. As they invoke applications on the network server and download and modify files, their local cache accumulates file blocks--the next time they invoke the application or download the file, the cache software only transmits data blocks that have changed.
Because much network access is repetitive, this means moving less data back and forth. Often these blocks are compressed to save additional bandwidth and blocks that are "near" the requested data are "pre-fetched," reducing latency. The net effect to the user is significantly faster response time, often a 200 percent to 300 percent improvement.
Network managers must bear in mind, however, that current network cache products work only for communications that consist of a file transfer, such as E-mail and word processing. Client/server applications do not benefit and Web access response time improvements would probably be sporadic at best.
As mentioned earlier, users of Shiva's LANRover RASes and switches have an easy path to Remote Access Caching when running Version 4.5 or later of Shiva's software.
Managers must bear in mind that PowerBurst is only available for Shiva LANRover RASes and switches, works only on IPX networks, and does not have Windows NT client cache software. Support for accessing TCP/IP networks will not be available until later this year.
Point B Remote Net-Accelerator looks promising for managers who must accelerate remote file downloads and uploads from IP and IPX networks. Point B includes client software for Windows 95, Windows 3.1 and Windows NT, and a server component that resides on an NT server on the network and can be independent of the network server.
By providing access from WANs, VPNs and dial-up lines, Point B offers the broadest support for remote users. Its caching routines speed file writes and reads when the user locally reconnects to the corporate network. Point B employs a Least Recently Used algorithm to clean up its cache, vs. PowerBurst's first-in, first-out approach.
TurboGold, an older remote access cache product, is a software-only solution that accelerates file downloads locally over a WAN or via a RAS from NT and NetWare networks, but turns off caching when the client detects that the network server is less than 100 ms away.
Enhanced transport control
Enhanced Transport Control rapidly finds and dynamically maintains a transmission rate close to the theoretical capacity of the channel. All data transmitted via TCP/IP benefits from the speedup, including bi-directional file transfers and client/server communication.
runTCP is an IP-compatible device that provides Enhanced Transport Control and is installed between a corporate network or an ISP's server farm and a WAN router or Internet connection point. runTCP acts like a session-level repeater, buffering incoming data locally until receiving an acknowledgment from the local client. It has two 10BaseT ports, one connects to the hub, the other to the router, and requires no other changes or further management.
This can be extremely beneficial when the remote data source is far away and accessed over a costly link, because retransmissions are then local, not across the WAN. Remote users downloading from the corporate network generate incoming traffic in the form of acknowledgments that compete for bandwidth with incoming data to the network.runTCP smooths this traffic, reducing cross-load degradation and improving performance for both classes of users. When RUN first developed this technology (for AppleTalk networks), it achieved an actual bandwidth of 1M bps, compared with the theoretical limit of 1.12M bps and the typical 200K bps. runTCP requires no management; network managers can just set it and forget it.
If the goal is to efficiently exploit current bandwidth, then a straightforward approach is to create and enforce a bandwidth policy for all users, applications, and data sources on the network that fits within available resources.
Bandwidth Partitioning devices bring to IP features similar to the minimum and maximum quality-of-service guarantees of ATM and the resource reservation of RSVP. Managing the use of a pipe--and not exceeding its actual capacity--can reduce or eliminate dropped packets that must be retransmitted with additional acknowledgments, improving efficiency and effective bandwidth.
Aponet and Xedia implement a queuing approach to enforcing bandwidth constraints (Xedia exploits advanced Class Based Queuing), whereas Packeteer's PacketShaper implements a rate-based control by dynamically adjusting transmit and receive windows, effective even when a slow pipe, such as a T-1, dumps into a faster network, such as a corporate intranet.
Policies can be set up to support both maximums and burstable minimums. Users and testers of PacketShaper have found no incompatibilities with current TCP stacks. Aponet's focus is on bandwidth allocation and management, not specifically on performance improvements, therefore it provides extensive monitoring and reporting functions. Xedia now supports 100M-bps technology, and Aponet expects to ship products supporting it in the fourth quarter. Packeteer also plans support in the near future.
PacketShaper smooths and speeds transmission by effectively modifying how it "pulls" packets, adjusting the receive window it returns to the sender. runTCP smooths and speeds transmission by quickly "pushing" packets up to the fastest speed that doesn't incur packet loss. A combination of the two could boost performance by five times or more, with runTCP providing greater effective bandwidth for PacketShaper to utilize.
If such combinations prove their worthiness, then an all-in-one box would be a very desirable commodity.
Erick Von Schweber is a principal of Infomaniacs, a think tank in Sedona, Ariz., specializing in technology convergence. He can be reached at firstname.lastname@example.org or www.infomaniacs.com.
Contact the authors send mail to
Infomaniacs home is www.infomaniacs.com
Updated Jan 28, 1998