Disk versus Drum Drums and head per track disks, as well as most semiconductor in Software

Creation QR Code JIS X 0510 in Software Disk versus Drum Drums and head per track disks, as well as most semiconductor

memories, do not require a seek When the seek time s is zero, the e ect of rotational latency and transfer rate becomes more important than the discussions imply All the formulas developed will still hold when s = 0 The bene ts of double bu ering is relatively greater, since it applies to the latency

Sec 2-5

The methods discussed also apply to the use of tape for les The block size B for tape is speci ed by the system used, and can be used also where the track size is required Most tape units can read tapes in both the forward and the reverse direction and this provides the e ective random seek time Writing may only be possible in the forward direction The expected random seek time for reading s can be estimated based on forward or backward searching through a third of the length of tape in use Even when using the ability to read in either direction, this operation will take more than 1 min For writing, or if reverse reading is not possible, other values for the seek times will have to be determined, depending on the method used to nd a record A simple approach is to always rewind the tape and then search forward Better strategies can be used if the current position of the tape is remembered and used This permits tape to be spaced forward or backward prior to forward search to achieve minimal seek delays The latency r is due to the time taken to accelerate the tape to full speed The gap G provides the space for getting up to speed and slowing down The extremely high value of the expected random seek times e ectively removes tape as a candidate for the le organizations presented in Chaps 4 and 5 Some operations can be faster on tape than on disk To nd a predecessor block on a disk unit takes most of a revolution On a tape unit with reverse reading such a block can be read immediately in reverse mode Even without reverse reading it is possible on tape to backspace over the block and then read or write it forward This assumption was made for the rewrite time in Sec 2-3-5 We can incorporate the e ect of rewrites for tape to the formulas of Chap 3 by setting the value of the latency r to the block transfer time for tape Similarly the single cylinder seek s1 can be kept as 2r These simpli cations introduce errors in the value of s for the seek times, but the e ect is negligible because of the long seek times for tape under random access conditions More complex le architectures, as introduced below, require further analysis The principles underlying the parameter derivations can be applied also in those cases On unusual devices it will be wise to run some tests, comparing measurements with analysis results Avoiding analyses and only relying on measurements does not help in understanding, and can easily fail to reveal what happens in extreme or boundary conditions

Storage is a major component of a computer system, as indicated in Fig 1-1 Storage has to interact with all othe components of a computer system, and performance gains can be achieved by improving the overall interaction In this section we review some of the architectural alternatives that have been considered Table 2-6 clari es the role of storage in an overall computer system hierarchy

Table 2-6

Processor registers for arithmetic Cache memory for high-rate computation Main memory for program and data, and IO bu ers Small, rapid disks with multiple heads for working storage Main data storage, consisting of multiple large moving head disks On-line archival storage, perhaps using optical discs or tape cassettes Archival o -line storage, often using reels of tape, kept remotely for protection Architectural arrangements that have been used include 1 Back-end conventional processors to o oad heavily loaded central computers Some database management systems have the option that all transactions addressing intensively used but limited size les are assigned to a distinct processor which can keep all or most of those les in memory 2 File servers, composed of conventional processors con gured to only provide le services These are especially common in distributed environments, since they permit users to select any of a number of connected processors for their computation 3 Back-end specialized processors which perform scanning and simple primary key access to rapidly fetch selected data Performance gains of a factor of 2 or 3 are typical 4 Systems using multiple processors dedicated to one or a few storage devices They may interpret arbitrary retrieval and update requests for the les under their control 5 Computers using associative memory for rapid analysis of data blocks, used in specialized applications such as image data processing and sensor data reduction We now describe in more detail several architetural options 2-5-1 Database Machines Database machines use specialized hardware to improve the performance of operations using external storage Most e orts concentrate on making le access faster by placing important operations into hardware An important by-product is the modularity of systems which is needed to achieve such a partitioning Opportunities for performance improvement exist at all stages in the ow of data in a computer system: 1 Faster disk access by parallel access to data on disk 2 Faster data transfer by early selection and projection of relevant data 3 Faster operations by hardware implementation of operations 4 Reduced competition for the processor by performing the data reduction in a separate back-end computer 5 Better tracking of modern technology by being able to improve back-end or main processors independently, as it becomes advantageous Some conditions must be satis ed before the advantages will be realized

Sec 2-5