FRAMINGHAM (05/01/2000) - Our test environment consisted of six routed Fast Ethernet subnet domains and a T-1 connection to our ISP. The Internet link let us perform massive zone transfers and other large-scale IP address operations, but most of our testing occurred just on our network 's intranet. Throughout the subnet domains, we ran several concurrent instances of a C++ program we wrote that issued DHCP-DISCOVER messages. Some of these messages were valid requests for IP address information, but we also deliberately created many invalid requests. Our invalid situations included duplicate requests, missing DHCP-REQUEST messages and lease renewal requests at other than the usual lease half-life interval. To cause address reassignment, we forced clients to frequently join and leave the network. To test performance, we used a stopwatch to measure how quickly each Dynamic Host Configuration Protocol (DHCP) server could assign 50,000 IP addresses.
We also tested the products ' handling of implied source subnet qualifier overrides to verify selection of the correct address for a particular logical network. This test helps us assess how a product behaves when you need to migrate multiple subnets on a single broadcast segment to a flat, switched network topology.
We moved clients from one subnet to another, gave unique values to the DHCP client ID field and assigned different values to the user class ID and vendor class ID DHCP parameters to see how the DHCP servers would respond. We looked at how each vendor 's DHCP server handled sequences of valid and invalid DHCP-DISCOVER, DHCP-OFFER, DHCP-REQUEST and DHCP-PACK messages. We tested each DHCP server 's ability to respond to a variety of different client platform 's TCP/IP protocol stacks with DHCP configuration options appropriate to those platforms. The 25 client computers on our network were a mix of Windows 2000 Professional, NT Workstation 4.0, Windows 98, OS/2 Warp 4.0 and Macintosh System 8 platforms.
We also evaluated each product 's dynamic Domain Name System functions as well as its ability to efficiently update its DNS database with names and addresses from the ISP 's DNS server. We even combined these products in various ways to determine their interoperability.
We factored in the ease with which we could administer the products, centrally or remotely, paying particular attention to each vendor 's address pool maintenance and network monitoring tools. We looked at each product 's support for BIND 8 and DHCP Version 6; determined each product 's level of security; and took into consideration what each vendor offers by way of scalability and fault tolerance.
Except for Network TeleSystems ' Shadow IPserver, which is a combination hardware and software product consisting of two network appliances preloaded with the IP address management software, we ran the address management software products on three Gateway NS-8000 computers with 333-MHz Pentium II dual processors, 512M bytes of RAM and three 9G-byte SCSI RAID drives. In each case, the operating system platform was Windows NT Server 4.0 with Service Pack 5.
For reporting and querying tools, we used either the database that shipped with the product or one recommended by the vendor to store the IP address data.
Network Associates Inc.'s Sniffer protocol analyzer software, running on a Dolch PAC63 computer, decoded and displayed network traffic.