![]() |
|||
![]()
|
![]() |
![]() Click Here! |
![]() |
Measured System Sensitivities of Typical Microcomputers The previous section reviewed the theoretical sensitivities of microcomputers at both the chip and system levels. The chip level sensitivities are hard facts. They represent the physical characteristics of semiconductor devices, tolerances published by designers and vendors of high-density integrated circuits. System sensitivities can never be described as precisely. At the system level, the design and environmental variables are hard enough to measure in a laboratory setting let alone predict over the full dynamic range present in the real world. It is generally accepted that capacitive coupling is the mechanism by which transient voltage disturbances are transferred to a computers internal components. Even so, many experts in the fields of computer system design and power protection design disagree on just how sensitive systems are to various types of transient electrical events. The basis of the disagreement centers on assumptions regarding the filtering and attenuation properties of power supplies and other in-system, DC circuit protection schemes. During 1992, research on system-level sensitivity to electrical phenomena was published by PowerCET Corp. (Santa Clara CA), an independent consulting organization. The study compared the electrical sensitivities of two microcomputers, one a well-designed model from a major computer vendor and the other a cost-leader 286 IBM clone. Summary of PowerCET Research PowerCETs research demonstrated in a controlled laboratory environment that the well-designed microcomputer was less sensitive to electrical-power events than the budget machine, but on both computers interference currents were measured on the +5-V DC bus that supplies the system mother board. These interference currents caused functional problems in both machines, even when the external noise voltage was under 50 V. The report concluded that system sensitivity is a function of:
Symptoms of Electrical Interference PowerCETs research demonstrated that no commercially available computer hardware is completely immune to electrical interference. Electrical interference can cause spurious hardware interrupts leading to abends in operating systems (e.g., NetWare or OS/2) that typically run in protected mode. Novells engineers recognize the consequences of hardware susceptibilities to electrical transients. In the system messages manual for NetWare 2.2, poor power line conditioning is cited as a probable cause of GPI, NMI, and at least 15 other system errors. Novells AppNotes on 386 NetWare System Messages say, The majority of NetWare operating system messages are of the fatal/abend type. Fatal/Abend messages are usually caused by consistency checks. Not all consistency check errors are caused by software anomalies. These errors might also be related to corrupted OS files, defective memory chips, static discharges, faulty power supplies or power surges and spikes. This authors experiences with trouble-prone NetWare installations suggest a strong correlation between fatal/abend occurrences and the presence of electrical noise. The following is a representative scenario. Case Study Situation A team of consultants was working to develop a client/server application with a graphic user interface that would automate processing of home and auto insurance applications in regional offices across the country. The system was being implemented on a NetWare 3.11 platform running on an IBM 8595 server. A prototype system was scheduled for rollout. One of the critical milestones that needed to be reached before moving ahead with development was to verify the stability of the system. Problem Random hardware interrupt errors that crashed the server occurred four times in a three-week period. These unexplained system faults caused delays in development and were a source of much frustration. Literally, tens of consulting hours had been spent trying to identify the cause of the problem. The systems development team assumed that the power supply was satisfactory. The electrical supply in the development lab consisted of dedicated, isolated-ground circuits with surge-suppressed outlets. The server was powered by a 900-volt-ampere smart UPS of a suppressor/ filter type architecture. A system engineer from Novell was brought in on the problem. Based on the nature of the hardware interrupt symptoms, the system engineer suspected the cause to be related to power and grounding anomalies. Diagnosis An evaluation of the electrical service supplying the server room revealed no irregularities. All circuits were correctly wired and nominal voltage levels were between 115 V and 122 V. A power quality recording instrument was installed to measure and record the quality of power of the server. Transient impulses of 70 V to 80 V were recorded during the period that the system reported a series of spurious hardware interrupt errors that the system engineer from Novell believed were related to power problems. Solution The existing suppressor-type UPS was replaced with a UPS that incorporated full output transformer isolation and high-frequency filtering circuits. Random hardware interrupts no longer occurred. System stability was demonstrated and the project was put back on schedule.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |