Security
W O R K S H O P  
For VPN Servers, Scalability and Throughput are Critical

  May 14, 2001
  By Tom Zeller


If you are responsible for your company's network services, your to-do list probably has an entry for VPN. There's a good reason for the explosion in the market for VPN servers: A VPN server provides a secure gateway for letting your company's employees access your corporate LAN remotely via the Internet. A VPN server eliminates long-distance charges for dialing directly into the corporate network and provides a secure link between business sites over an existing Internet link.



Security is at the core of the VPN. When properly configured, the VPN server will pass packets only from users who have been authenticated. The server encrypts the information from the corporate servers as it is sent over the Internet to the remote user. The remote user appears to be a local user, permitting access to services restricted by IP address.

The Primary Design Factor: Scale

When you look at the abundance of VPN servers out there, the criterion of scale will let you zero in quickly on the subset of servers that will meet your needs. Two related parameters will help you sort through the contenders: the number of simultaneous users and throughput capacity.

Numerous VPN solutions are sized for 100 or fewer users. If your environment demands a larger scale, the number of choices drops dramatically, and the prices rise just as dramatically. On the high end, VPN servers targeted at Internet service providers can handle 100,000 simultaneous users.

Another dimension of scale for a VPN server is the rate at which it can process packets and shove bits down the pipe. Each packet received must be compared against any input and output filters for address, protocol type, port number and other parameters. If the packet passes the filtering process, the payload must be encrypted or decrypted, depending on its direction through the VPN tunnel.

Because the encryption process is packet-oriented, performance will vary depending on the size of the packets. Processing a megabyte of 1,000-byte packets requires about 1,000 encryption operations. The same megabyte arriving in 100-byte packets requires 10 times the number of operations. As a result, the raw performance in bytes per second for a series of 1,500-byte packets will be several times that for a stream of 300-byte packets, the average packet size on the Internet. Published specifications typically refer to the results for large packet sizes, so real-world mileage will vary.



Another variable is the performance degradation some systems experience because of packet fragmentation from the insertion of additional IPsec header information. (For information on VPN performance testing, see "IPsec VPNs: Progress Slow But Steady".)

The relationship between raw byte processing and the number of users supported must also be considered. Some VPN servers rated to support 100 users can process only 1 million to 2 million bps. Do the math. That's only 10 Kbps to 20 Kbps per user if all the users are sending packets at the same instant. If they are all watching video streams, there will be contention for server bandwidth; packets will be dropped, and the users will experience poor performance. If, on the other hand, they are all casually browsing Web pages and reading e-mail, they may find the experience acceptable. Other 100-user systems support higher throughputs, up to about 10 Mbps. Note that an encrypted frame is near-random and will not benefit from compression.

Servers that support several hundred to about 1,500 users have throughputs on the order of 20 Mbps to 80 Mbps, but such systems can't be bought for mere milk money. One reason that higher-speed VPN servers get so pricey is that they use expensive, specialized hardware to perform the packet encryption/decryption function.

When IPsec is the VPN protocol used, very large installations have to pay attention to the number of IPsec connections that can be established per second. Quite a bit of CPU activity is involved in setting up an IPsec connection, and a server capable of handling thousands of active connections may be able to create only a few hundred in a second.

The final parameter of scalability is capacity expandability. Precisely determining the number of users your system will be required to service at any given instant may be difficult. If you misjudge, you'll want a pathway to higher service levels. That may not be necessary, however, if you think you can realize sufficient return on investment on your first server in two to three years and buy a bigger server when you need it.

Some servers are purchased with licenses for a certain number of users. Later you can buy more licenses, but the hardware you're using must be able to support the additional users. Some servers provide a hardware upgrade path. This might simply mean adding one or more CPUs to the server to aid it in the encryption process. Other systems allow the installation of additional encryption/decryption hardware in the field to increase throughput.

Scalability can mean a multitude of VPN servers in different locations. If you're going to support many VPN servers, consider the management applications some vendors offer that allow a single point of management for many servers.


   Page: 1 | 2 | Next Page

Research and Reports

Storage Virtualization Guide
May 2012

Network Computing: May 2012

TechWeb Careers