"There are a number of areas in which NetWare 4.1 stands alone, or in very limited company. As for core services, only Banyan VINES comes close to matching NetWare 4.1's built-in Directory Services and integrated messaging, but its implementations are not as powerful, scalable, or available to as many clients. In security and network management, NetWare 4.1 clearly outmatches the others. Where multiprotocol routing is concerned, NetWare stands alone as the only NOS that incorporates complete multiprotocol routing in its base offering. In Enhanced Services, the depth of NetWare 4.1 is clear compared to the other NOSs. Only in remote access do the other systems offer similar functionality with NetWare's. In all other areas, their functionality is limited at best. In the areas of telephony, imaging, document management, and storage management, NetWare 4.1 has literally no competition from the NOSs surveyed here.
The message should be clear: NetWare 4.1 offers stronger core services to its users than the competition, and in the area of enhanced services offers more and better offerings than any other NOS.
PERFORMANCE COMPARISON
The preceding section should make it clear that as an NOS, NetWare 4.1 stands in a feature-rich class of its own. This section compares the performance of the other network operating systems; including Windows NT 3.5 and IBM LAN Server 4.0, and shows that NetWare 4.1's performance also puts it at the head of the class.
Novell conducted performance tests in its Provo, UT, SuperLab to benchmark these network operating systems in one of the largest and best-equipped network testing facilities in the world. Its findings shouldn't be surprising: these tests show conclusively that NetWare 4.1 remains king of the performance hill.
To be as fair and objective as possible, the tests performed in the SuperLab used the recently-introduced NetBench 3.0 benchmark test suite developed by the Ziff-Davis Benchmark Operation (ZDBOp). Released in September, 1994, NetBench 3.0 is specifically designed to test network server performance. This test suite includes the following testing tools: Various Read/Write mixes, a network interface card (NIC) test, and a new Disk Mix test.
TEST METHODOLOGY
To eliminate any hardware specific differences, all the NOSs were run on the same server, with the following configurations:
Compaq Proliant 4000
128 MB RAM
1 Pentium 66 MHZ CPU
4 2GB disks (in a Compaq disk array)
4 Netflex II Ethernet controllers (for one set of tests)
4 NE 3200 Ethernet controllers (for another set of tests)
The clients used consisted of a variety of 486, and Pentium-based workstations, running DOS and MS Windows 3.11, with various applications. Novell's testing was conducted with strict care to optimize client and server configurations for all three NOSs and for all clients. The idea was to measure real system throughput fairly and accurately for all contenders.
However, using synthetic benchmarks like NetBench 3.0, which only simulates application activity without actually running real applications, to predict system performance can be confusing and sometimes misleading. Consider, for instance, what happened with the Read/Write mix test run on NetWare 4.1 and MS Windows NT Server 3.5.
The two systems appeared to perform fairly equally. But what actually takes place is that given this particular workload, both systems can consume the entire bandwidth of the network medium (Ethernet). A more relevant metric for this scenario is the CPU utilization for each system during the test. This indicates how much computing power is needed by each system to fully utilize the medium's carrying capacity.
This measurement is important for several reasons: First, it indicates a system's hardware requirements (how much CPU power it really needs). Second, it indicates how efficient the system really is. And finally, it predicts fairly well how an operating system will scale. For instance, if the medium were replaced by a faster technology like FDDI, or 100-Megabit Ethernet NT Server might require a second CPU, or a third, or perhaps even a very costly fourth CPU, to take full advantage of the expanded medium. This, however, is not the case with NetWare: During this test its capabilities were not even stressed (and NetWare is already in use with these fast media in production environments today, using only a single CPU). To be more specific Novell observed during the Read/Write tests that NetWare 4.1's efficiency allowed it to saturate the Ethernet, while only utilizing half of the CPU compared to Windows NT Server 3.5. Averages measured 45 percent for NetWare 4.1 vs. 91 percent for NT Server 3.5 .
More important, the results of this artificial test can be observed in NetWare as it performs in a real-world situation. Sites using NetWare typical measure server utilization in the 25 to 45 per cent range, whereas the medium will be 55 to 60 percent saturated. This means that as new transport mechanisms are introduced, NetWare can use their increased bandwidth without requiring further server augmentation. NetWare's lean, yet feature-rich architecture explains why NetWare has remained the performance leader.
MEASURING THROUGHPUT
According to the NetBench 3.0 Manual, the goal behind the Disk Mix test is to 'enable you to measure in a realistic way the speed of your server as it handles file operations.' It does this by ' taking eight leading PC applications and profiling them to determine what sort of file requests they performed and how frequently they performed them. The profiling tools captured the entire network file behavior of the applications, this research data indicates how applications distribute operations, not just what the operations are.'
After profiling the operations, ZDBOp implemented this workload using a scripted set of system operations, called the Disk Mix, thereby emulating a mix of normal networked applications to model input-output. Each client randomly orders these operations to insure that different disk mix components run at different times to mimic the randomness of the real-world. Disk Mix models the following applications' behavior:
Lotus 1-2-3
Microsoft C compiler
Borland's Dbase IV
Borland Paradox for DOS
WordPerfect for DOS
Harvard Graphics
Lotus CC:Mail for DOS
Microsoft MAIL for DOS
Each client in the test creates a sizable workspace (default size: 20 MB). Creating such a large test area insures that the clients have to read from the entire disk, rather than from a single disk region, or simply from cache. Also, because the disk sub-system is such a vital part of Disk Mix, Novell tried to test the operating systems in as many different configurations as it could, using different controllers, different disk striping arrangements, etc.
During the Disk Mix test Novell evenly distributed the clients across four Ethernet segments. Starting with one client per segment, they increased the number of clients per segment equally until a total of 64 clients was reached (16 clients per segment). Here again, NetWare 4.1's performance typically outstrips that of Windows NT Server or LAN Server.
UNDERSTANDING BENCHMARK RESULTS
Novell's own benchmarking illustrates what customers have been telling us all along. Because NetWare can saturate the network medium at a significantly lower level of CPU utilization than its competitors, it demonstrates a more efficient usage of hardware and system resources. This in turn points to NetWare's superior scalability: Because NetWare 4.1's efficiency provides more power to exploit faster networking media as they are deployed on more and more networks, organizations should continue to be able to use existing servers and configurations with those media without requiring expensive hardware upgrades. NetWare 4.1's scalability makes it an ideal choice for tomorrow's higher-bandwidth networks.
Overall, NetWare 4.1 also demonstrates superior throughput, when compared to other network operating systems. Novell suggests that this may be the result of ten generations' worth of tuning and performance enhancements, as well as a system design that has stressed maximum server accessibility and response time from its very outset. These results clearly demonstrate that all of these factors combine to position NetWare 4.1 as a veritable speed demon when it comes to providing network services."
