In terms of the analysing process the best measure of secondary storage performance would be one that records the time taken for the actual data required by a particular analysing process to be read and successfully placed into RAM. Unfortunately trying out different models of hard disk for a particular scenario is generally impractical.
Also, in the real world most hard disks are used to store data supporting many different applications, which include many different types of analysing processes. For example, hard disks on a file server need to move quickly from reading one area on the disk surface to another as different users retrieve different files; seek and latency times measure such performance. Seek time is the average time taken for the read head to move in or out to a given track and then latency time is the average wait time for the particular data to arrive under the head.
Other analysis processes require a single large file to be retrieved. For example, a graphic designer is likely to retrieve single large sized image files, the critical requirement in this case is the speed of data transfer; the time taken to locate the file being relatively insignificant.
Measures to determine speed of data transfer include spindle speed and areal density. Spindle speed is the speed at which the disk rotates and is commonly expressed in revolutions per minute (rpm). Higher spindle speeds mean more data passes under the read head in a given time period, hence higher data transfer speeds. Note that higher spindle speeds also improve latency
times. Areal density is a measure of the maximum number of bits that can be stored on each square inch of the disk surface. In general, higher areal density means more data passes under the heads and hence the higher the data transfer speed will be. The incredible increases in hard drive capacities and data transfer speeds is largely a result of increases in areal density together with the technology to read such tightly packed data.
Most hard disks include a fast memory area called cache; during read operations data passes into cache, this includes the required data together with data the system predicts may soon be needed. Cache is fast chip-based storage; in a hard disk it is included on the hard disk’s circuit board. Data within the hard drive’s cache can be retrieved many times faster than data on the actual hard disk. If the prediction is accurate, meaning the required data is found to be in cache, then access times will be considerably faster. For example, the hard disk on the machine used to write this book recorded a read access speed of approximately 26 MB per second when none of the data resides in cache, however if all the data is currently in cache then the read access speed is closer to 370 MB per second. Therefore, the amount of cache contained within a hard drive can have a significant effect on data access speeds.
Fig 5.3
Internal view of a hard disk. The spinning disk platters and read/write
head arms can be clearly seen.
RAM (RANDOM ACCESS MEMORY)
In terms of the analysing process the total amount of RAM installed is the most critical measure. RAM holds both the software and the data used by the CPU during processing. If there is insufficient space available in RAM then required instructions or data must be repeatedly written to and retrieved from secondary storage. As secondary storage is many thousands of times slower than RAM, a noticeable drop in performance will certainly result. To put this into perspective, a retrieving process that would take seconds using RAM will take hours using a typical hard disk. Often the cheapest and most effective means for improving performance is to add extra RAM. At the time of writing most personal computers contain a minimum of 128MB of RAM and many contain as much as 1Gb; it is likely that these figures will continue to increase.
The speed at which data held in RAM can be accessed is important for analysing processes; different types of RAM are able to operate at different speeds. The speed at which RAM chips are able to
deliver data is determined by the amount of data transferred as a single unit together with the speed at which data can be stored and retrieved. Both these factors must correspond to the specifications of the CPU and also the motherboard onto which the RAM module and CPU are installed.
CPU (CENTRAL PROCESSING UNIT)
The number of bits that can be processed simultaneously, the speed at which instructions are executed and the nature of the instructions are just some of the factors determining the performance of the CPU; there are many others. Different CPU designs are suited to different types of processes. If all other
specifications are equal then a CPU capable of processing 64 bits at a time will be twice as fast as one that processes 32 bits; similarly a CPU with a clock speed of 2GHz processes at twice the speed of one with a clock speed of 1GHz. Such measures are only reliable for processors of the same design.
Different CPU designs use different sets of instructions and different techniques for executing these instructions.
Executing a particular analysing process, such as averaging a set of numbers, will likely require quite a different number
of clock cycles on different CPU designs. The instruction set for each family of CPU is different therefore on different CPUs a given process is likely to require the
GROUP TASK Activity
There are numerous software utilities available for analysing the
performance of hard disks. Download such a utility from the Internet, or otherwise, and use it to assess the performance of your hard disk.
Fig 5.4 DDR-RAM module containing 256MB
of memory.
GROUP TASK Research
Research, using the Internet, different types of RAM module. Classify each according to the amount of data transferred as a unit, together with the speed of transfer.
Fig 5.5
An Intel Pentium 4 central processing unit.
execution of a different number of CPU instructions. Furthermore, different processor designs are able to execute a different number of instructions at the same time, these instructions being executed in parallel and also being at different stages of execution.
It seems that comparing the performance of different CPUs is an impossible task.
So how can we determine the best CPU for a particular information system’s analysing processes? As was the case with secondary storage, the best method is to execute the same real analysing processes on various types of CPU and record the time taken. This is not practical for all but the largest systems and furthermore the CPU does not operate in isolation, hence other hardware components will affect the outcomes. There are various companies, including various magazines that perform benchmark tests designed specifically for this purpose. Although such tests are unlikely to replicate precise analysing tasks, at least they do provide results that are unbiased. Often these benchmark tests are performed using a variety of software and data scenarios, this makes it possible to select a scenario that best emulates the type of analysing processes relevant to the current information system.
Consider the following:
• If the total amount of data is such that it can be held in RAM during analysis then the speed of the CPU and data retrieval from secondary storage is really not significant.
• For large stores of data it is impossible to retrieve all the data prior to analysis;
hence the access speed of secondary storage hardware is critical.
• If the hard disk lights on a file server are continually flashing then that’s a good indicator that more RAM is required.
• The most valid means of comparing the performance of different CPUs is to consider their clock speed together with the number of bits processed at any one time.
• Increasing the amount of RAM is the most cost effective means of improving the processing performance of most computer systems.
• Faster access to secondary storage devices results in higher performance compared to increasing the amount of RAM or installing a faster CPU.
• Creating large digital videos is an intensive CPU process, hence upgrading to a faster CPU would be the best method of improving performance.
GROUP TASK Discussion
Each of the above statements is partially correct and partially incorrect, it depends on the individual processes taking place. Describe scenarios where each statement is true and scenarios where each statement is false.
GROUP TASK Discussion
“Hardware should be selected according to the individual needs of each particular information system. Often this is just not possible, so
compromises are made.” Why is it that compromises are so often made?
Discuss.