PC performans testleri

PC performance tests

Tez No: 19251
Yazar: ERCAN ÜÇ
Danışmanlar: DOÇ.DR. BÜLENT ÖRENCİK
Tez Türü: Yüksek Lisans
Konular: Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Computer Engineering and Computer Science and Control
Anahtar Kelimeler: Belirtilmemiş.
Yıl: 1991
Dil: Türkçe
Üniversite: İstanbul Teknik Üniversitesi
Enstitü: Fen Bilimleri Enstitüsü
Ana Bilim Dalı: Belirtilmemiş.
Bilim Dalı: Belirtilmemiş.
Sayfa Sayısı: 112

Özet

Özet (Çeviri)

ÖZET Bu tezin amacı DOS işletim sisteminde çalışan bilgisayar sistemlerinin performans ölçümleri yaparak, performans açısından bu sistemleri kendi aralarında karşılaştırabilecek değerler elde etmektir. Kullanılan test programları bir standard olabilecek şekilde günümüzde BYTE dergisi tarafından kullanılmakta olan performans testleri temel alınarak geliştirilmiştir. Bu testler oluşturulurken bir performans testinin sahip olması gereken dört unsurun da bulunmasına dikkat edilmiştir: doğru olması, anlamlı' olması, tekrarlanabilir olması ve benzer sistemler için benzer sonuçlar verebilmesi. Tezde bilgisayar performansı alt düzey ve üst düzey olmak üzere iki grupta ele alınmıştır. Alt düzey performans testleri bellek, disk sistemi ve ekran sistemi testlerinden oluşmaktadır. Bu testlerde her bir alt sistemin performansı, standart bir bilgisayara göre bağıl olarak verilmektedir. Bellek testlerinde matris, sıralama, blok aktarım ve asal sayı bulma algoritmaları kullanılmaktadır. Bu algoritmalar farklı yönlerden MİB ve belleği test etmektedir. Disk üzerinde yapılan testlerde ise temel giriş/çıkış, işletim sistemi ve dosya düzeyinde testler uygulanmaktadır. Ekran testlerinde ise mevcut metin ve grafik modlarında ekrana bilgi yazılmakta ve ölçümler yapılmaktadır. Performans ölçümlerinin yanı sıra tezde, performansın arttırımı üzerinde bilgi verilmektedir. Bu amaçla mevcut disk sisteminden maksimum hızda yararlanılabilmesini sağlayacak geçmeli yapı ele alınmıştır. Tezin bir bölümünü oluşturan geçmeli yapıda program, diskin optimum hızda çalışmasını sağlayabilecek durumu belirleyerek, diski bu çalışma durumuna getirmek için formatlamaktadır. Bu işlem sırasında diskteki bilgilere zarar verilmemektedir. viYou can divide benchmarks into two categories: microscopic, looking at the components of a system in detail, and macroscopic, looking at the system as a whole. You must be careful not to interpret the results of macroscopic test as having meaning at the macroscopic level. Micro benchmarks are useful for finding the maximum capability of a component within a system. They are helpful in system design and in estimating maximum performance possible for an application under development. Choosing the appropriate benchmark requires some understanding of each benchmark proses and its relative accuracy. You may need to separate those benchmarks that test components from those that test systems, and try not to compare the results of the different categories of tests. And you need run only those tests that reflect the environment and work you indent to perform. If you add benchmarks beyond this, you increase work and confuse the issue by adding results not relevant to you objective. Having run a suite of benchmarks, how do you determine what they mean? There is no simple answer to this, because it depends on your objective. If you are evaluating processors for pure performance and use an appropriate benchmark, a significant difference between results makes a decision fairly easy. What makes a difference significant is the issue of repeatability. If you run the same test 10 times, are all the answers within 1 percent of each other? Within 10 percent? If you know that noise number, you can conclude the differences between system less than the noise amount are insignificant; a benchmark that shows differences between system less than the noise amount can't be used to differentiate between them. When doing the precise timings required of micro benchmarks, using a stopwatch is not good enough. The errors in start and stop times with a stop watch are on the order of milliseconds, while you want benchmark timings for micro benchmarks to be accurate to milliseconds if possible. The best way to time a micro benchmark is to the program use the system-timing functions to start and stop its“stopwatch”. When you need very precise timing, you must also estimate the time required to call the stopwatch functions. The incremental timing approach takes this time into account. With the exception of the memory trash benchmark, all the program in this project used a set of timer routines that should be relatively easy to port to any machine with timing functions. The benchmarks measure some subsystems of a computer system, not a whole system. On this project there are viiithree main subsystems tested: CPU and memory, disk and video subsystems. CPU and memory benchmarks are designed to test performance of memory unit using four testing methods: matrix, string move, sort and sieve. The results of each tests are reported separately. Sieve of Eratosthenes is used as a basic performance test by language and the system designers. The sieve makes use of arrays (indirect addressing), comparison operations, and simple math. It finds a fixed amount of primes using shortest way. The sort benchmark has three known sort algorithms: quick sort, shell sort, and heap sort. All three sorting techniques make heavy use of indirect addressing, comparison operations, and basic integer math operations. The differences of these sorting techniques are, quick sort is a recursive sorting procedure, while shell sort uses a repeating- loop algorithm, and the heap sort uses a repetitive call to a subroutine. The job of matrix benchmark is to exercise the kind of operations that must take place to manipulate matrices - in this case two dimensional arrays -. Primarily, this involves indirect addressing, but the Matrix benchmark also tests integer math operations. Note that some integer math operations must occur as a program calculates the offset of an array element. The Matrix benchmark makes three basic matrix operations: Add two square matrices, multiply two square matrices, and perform a transposition on a square matrix. In string move benchmark lots of bytes are moved from one place to another. String move benchmark uses assembly language routines. The benchmark moves bytes, a byte at a time and a word at a time. For the word at a time moves it tests in two ways. First, for odd-byte alignment and then for even-byte alignment. As storage subsystems become responsible for more of the time you spend at the computer, it becomes more important to get an accurate picture of how particular disk will affect total system performance.. If hardware and software vendors could agree on a single way to access disks, the job of benchmarking would be easy. Unfortunately for those of us who have to write and run the benchmarks, there are almost as many schemes for getting at storage as there are storage devices. In addition to the number of access methods and interfaces available, the number of components involved with any storage access makes accurate performance testing of any one component very difficult. In general, the storage device itself, the device controller, and the computer's operating system are involved with any ixtransaction. In some interface systems, such as small- computer- system- interface (SCSI) systems, a device driver is involved as well. In most situations, it's impossible and nonsensical to divorce the performance of the disk from its controller and driver. In this project it is not attempted to force this separation, concentrating instead on factoring various degrees of operating system overhead out of the performance equation. Since different programmers and hardware designers view the facilities of operating systems with varying degrees of contempt, it is varied the tests based on how heavily they rely on each operating system. On this thesis, harddisk benchmarks are held on in two categories: Basic Input/Output System (BIOS) level, and Operating System (OS) level. BIOS level benchmark manipulates the disk with the direct calls to the machine BIOS. This test is designed to gauge the speed of disk/controller combination, with as much of the operating system overhead as possible factored out of the results. The program begins with getting information from disk and then proceeds to test the disk in four stages, with each stage repeated 100 times. The program first seeks between the two outermost tracks, reading one sector on each. It then seeks between the two innermost tracks, the outermost and middle tracks, and the outermost and innermost tracks, reading one sector on each track. The times from all these seeks are collected, and an average seek time is calculated. The next benchmark is not nearly so subversive as the BIOS level test. In the disk subsystem benchmark, the operating system is used, but at a level of explicit device calls instead of relying on its file system. The testing methodology is relatively straightforward in an MS-DOS system. The main portion of the program is run three times. In the first iteration, the device is told to step through its sectors in 100 even steps, reading one sector in each step. In the second iteration, the same 100 step are used, but eight sectors are read at each stop, and in the third one,, reading sixteen sector at each step. The time for each of these iterations is reported. The file test program is designed to measure the interaction of the disk/controller and the operating system's file system. In the normal course of working with a storage device, three basic operations exists: creating files, appending files, and reading files. The benchmark creates 10 files of varying sizes. Each is then appended by a chosen amount. The test then uses an arbitrary chosen number ( actually a constant pseudo random number ) to determine location an size of thereads and writes that follow. Reading and writing in the last portion are performed at a ratio of three reads for every write, since this approximates the usage pattern of disk users that is observed. The accumulated times for reading and writing are returned, along with the total number of bytes written and read. The large file test uses all the help the operating system can offer to first write and read a 1 MB file. One of the most fundamental operations a microcomputer performs frequently is writing to its display. Faster display hardware permit the video operations to finish sooner, freeing the processor to handle other tasks. The first requirement for the test is that the effect of the software should be minimized. This means that the benchmarks should be coded in assembly language to minimize this effect an execute as fast as possible. However, another requirement of the benchmark, is that the benchmark programs have to run on a wide variety of machines. Assemly language procedures give the best possible speeds for a given microcomputer but would be useless on another machine and might not even work on the same machine with a new version of the operating system. These problems was solved by devising a set of simple low level video benchmark test that would test certain basic graphics operations. These basic graphics operations are: writing characters (text), positioning the cursor and manipulating pixels. All of tests discussed above are about subsystem level benchmarks. But, in general, what is important for PC users, is the whole system performance. So, how do we find the overall performance for a system?. First of all the overall performans depends on the applications used. For example, if it is a word processing program, file and memory subsystems are important. If it is a graphics program, the video subsystem is more important. Because of this dependency, on this project, the overall performance is giving for three main application groups: word processor, database and graphics programs. To get a result about application performans for each level, some micro benchmarks are run and the result is giving as a relative performance to the original IBM PC or AT, depending on the processor of the system tested. Then, for each group, under the heavy and light work, the results are giving. xi

Benzer Tezler

Tez No
892487
Mononükleer metalli ftalosiyanin komplekslerinin elektrokimyasal, eş zamanlı spektroelektrokimyasal ve elektrokatalitik özelliklerinin incelenmesi
Başlık çevirisi yok
LEVENT AYDIN
Yüksek Lisans
Türkçe
2024
Enerji Marmara Üniversitesi
Kimya Ana Bilim Dalı
DR. ÖĞR. ÜYESİ EFE BATURHAN ORMAN
Tez No
287478
GSM üzerinden PC tabanlı araç izleme sistemi tasarımı ve yapımı
Design and construction of a PC based vehicle tracking system via GSM
MUSTAFA BALCILAR
Yüksek Lisans
Türkçe
2010
Elektrik ve Elektronik Mühendisliği Gazi Üniversitesi
Elektronik-Bilgisayar Eğitimi Ana Bilim Dalı
YRD. DOÇ. DR. MUSTAFA BURUNKAYA
Tez No
533699
Tekstil endüstrisinde fonksiyonel polimerlerin yardımcı kimyasal olarak kullanımının incelenmesi
Investigation the use of functional polymers as auxiliary chemical in textile industry
HACER YEŞİM CENGİZ
Yüksek Lisans
Türkçe
2018
Kimya Mühendisliği İstanbul Üniversitesi-Cerrahpaşa
Kimya Mühendisliği Ana Bilim Dalı
PROF. DR. HÜSEYİN DELİGÖZ
Tez No
305763
An FPGA based BLDC motor control system
FPGA tabanlı fırçasız doğru akım motor kontrol sistemi
SERDAR UYGUR
Yüksek Lisans
İngilizce
2012
Elektrik ve Elektronik Mühendisliği Orta Doğu Teknik Üniversitesi
Elektrik ve Elektronik Mühendisliği Bölümü
PROF. DR. HASAN CENGİZ GÜRAN
Tez No
797766
Futbolcularda 8 haftalık mat pilates ve reformer pilates egzersizlerinin fiziksel parametrelere ve teknik beceriye etkisi
The effect of 8-WEEK mat pilates and reformer pilates exercises on physical parameters and technical skills in soccer players
OSMAN YILMAZ
Doktora
Türkçe
2023
Spor Selçuk Üniversitesi
Antrenörlük Eğitimi Ana Bilim Dalı
PROF. DR. TURGUT KAPLAN

Geri Dön