One can also easily see that no other component follows this growth rate.
Memory, RAM and storage are getting cheaper. However, what is not happening is that the performance of the access time and decrease resp. increasing at an exponential pace. Considering the technology of magnetic disk specifically we see that disk density has been improving by about 50% per year, almost quadrupling in three years. Access time has only improved by one-third in 10 years.
Super fast processors and huge memories have to be ‘fed’ and a system is only as fast as it’s slowest component, currently the disk.
In our analysis we shall consider the advantages and disadvantages of currently available ...view middle of the document...
Two basic types of benchmarks exist in the computer industry : those that run a real-world workload (ex.Winstone 97® and Winbench 97®), and those that simulate a workload (ex. ThreadMark). Each have their strength and weaknesses, and it often works well to run both types.
Real-world benchmarks may be as simple as timing how long it takes to copy a file, or they may involve executing complex scripts that perform functions in many different applications. The main advantage of these benchmarks is they measure performance using real applications.
this same advantage can also be a limitation, since the performance of one component may be artificially limited by another. For example, we may be interested in disk performance, but a slow video accelerator will slow down a benchmark that has a high proportion of video content. This in turn could affect the workload presented to the disk drives.
Why Benchmark results may not be what we expect
The single largest contributor to low performance is the disk drive mechanical overhead. The mechanical overhead consists of drive head seeks and rotational latencies. Rotational latency is the time it takes for a sector on the disk media to rotate to the point under the read/write head. It is inversely proportional to the rotational speed of the disk : the faster the disk spins, the shorter the rotational latency.
•Software and Firmware Overheads
Software overhead is the time it takes for commands to be passed through the operating system and software drivers. Firmware overhead relates to the time it takes for the host adapter and disk drive to process the commands. Both are substantially smaller than the mechanical overhead, and usually fall under 100 microseconds.
•Data Transfer Rates
Compared to the mechanical overheads, the data transfer periods are also very short. For example, it only takes approximately 200 microseconds to transfer a typical 4 KB block of data at the 20 MB/s Ultra SCSI rate.
To illustrate how the different overheads affect performance, consider this example. A good 7200 rpm hard drive will have an average seek time of 8 milliseconds, and a rotational latency of 4.2 milliseconds. The combined soft- and firmware overheads will only add 100 microseconds, and the data transfer time for 4 KB of data will be about 200 microseconds. All the factors together amount to 13.5 milliseconds of overhead. The soft- and firmware overheads together with the data transfer time only account for 2.4 % of the total overhead.
Drive media rates, spindle speeds, buffer sizes, system bus speeds and bus chipsets are important items that can have a major effect on the overall data transfer rate. The higher the drive’s media rate, the faster the data can be read into or written from the drive’s buffer. The drive’s media rate is closely tied to its spindle speed, since the faster the disk platters spin, the more data passes the drive heads in a given time. Today’s...