Novell NetWare Performance Note Hewlett-Packard Company General Systems Division 4/28/92 This note contains the results of a benchmark comparing an HP 9000 Model 837S running Novell NetWare for UNIX (formerly Portable NetWare) with an Intel-based 386 PC running Native NetWare. This performance information should assist you in selecting the right PC LAN server for your needs. SUMMARY For large numbers of users, to accommodate future growth, or for running other applications in addition to acting as a LAN server, the HP 9000/8X7S is the best solution. Its low level of CPU utilization at greater than 50 users and its scalability make it the preferable solution. For PC LAN servers with few users and no additional load from other applications, both the HP 9000/8X7S and the 386 PC can effectively handle the load. A 486 PC showed throughput comparable to the 386 PC. BENCHMARK The PC LAN benchmark was originally published in PC Magazine on May 29, 1990. While not ideal, it is the most widely available PC LAN benchmark. The benchmark sequentially and randomly reads and writes to a one megabyte file. The test is run using three different buffer sizes and various network loads. The benchmark reports the throughput that one client will experience while various network loads are simulated with load stations, each approximating ten to twenty users. As with any benchmark, it has limitations. The following are some of the problems associated with it. o This benchmark does not in any way represent or predict the performance that you might expect from your application. It shows the performance, under ideal conditions, of one particular configuration. o This particular benchmark was not run using the same version of NetWare on each system. While the 386 PC used NetWare version 3.11, the HP 9000 Model 837S used NetWare version 3.01. Running NetWare version 3.11, which will be available soon on the HP 9000/8X7S, will boost performance further. o 10 to 20 users are simulated on one PC load station. This does not accurately represent the load of 10 to 20 real users. With 10 PCs on a system, they will sometimes request services at the same time. With one PC simulating multiple users, the requests are serialized. This will cause a different load on the system. Also, the reported server performance varies greatly depending on the performance of the clients. o This benchmark only measures throughput. Response time is also important. While the average response time may be reasonable, individual response times could be very erratic. SYSTEM CONFIGURATIONS The HP 9000/837S under test was equipped with a floating point co- processor and ran NetWare for UNIX Version 3.01. The PC was a 25 Mhz Intel 386. There were up to 10 PCs acting as client load stations. These clients included Compaq 386s, Vectra 386s, and Vectra 286s. PERFORMANCE RESULTS There are two important performance considerations when selecting the right server. First, the server needs to have enough power to handle the current user load of the LAN. If the power is insufficient, users will experience poor response time from the server. Equally important is the CPU utilization. If the CPU is 100% utilized, any increase in the number of users will significantly degrade the response time. Also, other applications on the same system will not run with adequate performance. This is particularly important for customers who anticipate future growth or who may want to use their server to run other applications. Performance and capacity utilization results are shown in the graph on the next page. Keep in mind when viewing this graph that the 386 currently has two advantages that will be going away in the coming months. The PC was running Native NetWare which is historically 10% to 100% faster than NetWare for UNIX. Also, the PC was using NetWare version 3.11 while the HP 9000/837S was using the older version 3.01. While the 386 has slightly better performance, the HP 9000/837 has better CPU utilization. The 386, at four loads, has 27% free CPU and is declining rapidly (this was not tested past four loads.) The HP 9000/837S, on the other hand, is at 30% free CPU and has leveled off. Even at 10 loads (approximately 100 users), the HP 9000/837S still has significant capacity available for other processing or additional users. A 486 PC, while not shown on the graph, generally performed similarly to the 386 PC. At most user loads tested, their throughput was within 10% of each other. .PA RECOMMENDATION When selecting a PC LAN server, consider how you will be using it. In dedicated environments with a limited number of users and no expectations for future growth, either the 386 or an HP 9000/817S or larger can effectively manage the workload. In this environment, look at other key factors such as existing environment and type of LAN applications being used (peer-to-peer file sharing, wide-area networking, etc.) When the system will be doing more than supporting a limited number of users on a dedicated PC LAN server, the HP 9000/8X7S is the right choice. For lighter loads, the HP 9000/817S will be adequate. For heavier loads, you will need an HP 9000/837S or larger. The HP 9000/8X7S's lower rate of CPU utilization makes it ideal for concurrently running other applications in addition to acting as a LAN server. Also, if the server will be supporting a large number of users, the HP 9000/837S is the safest choice. This benchmark showed that while the PC supported 50 users successfully (5 load stations), the HP 9000/837S easily accommodated 100 users* (10 load stations) with capacity to spare. Several other factors favor the HP 9000/8X7S. It can accommodate significant future growth. With its extra capacity, as well as the ease of scaling up to larger systems, it provides significant room for growth over the long term. The HP 9000/8X7S also has the ability to run thousands of existing UNIX applications. Finally, these systems provide strong network and systems management capabilities as well as high levels of connectivity to legacy systems. * The PC was not tested at greater than 5 load stations.