Colleague Carnegie Mellon university, doctor Bjank Schroeder published his own study in the cases of failure for the system hard disks. Let us recall, for a while back the same problem were dedicated material, published by Google.
Specifically, work Schroeder was honored the reward FAST'07 best paper in the category academic scientific computer research.
By material came out about 100 thousand hard disks , including those installed in the clusters Los Alamos and super-computer center, and also in several providers Internet- services.
Here are some the key conclusions, made Schroeder, according to the results of analysis.
Were noted the extremely small difference in the percentage of failure from the system of hdd, equipped with interfaces SCSI, Fiber channel and SATA, otherwise speaking, between the disks industrial and consumer destination.
Child mortality of disks, in the opinion of researcher, is not the underestimated problem. In this the report is differed from the point of view Of google. The percentage of failure from system disk, according to Schroeder, grows with their period of service. Report assumes that child mortality disks is sufficiently correctly evaluated by producers, whereas the early wear of hard disks, i.e., a considerable increase in the breakdowns after the year of work is in reality basic problem. At present standard IDEMA (International disk drive equipment and Materials association) provided the standards of failure for the disks those worked out 1-3, 4-6, 7-12 and 13-60 months. The in force standard, in the opinion of researcher, does not consider a constant increase in the number of breakdowns, beginning from the second year of service disk, and it needs modification.
The average yearly percentage of hdd failure varies in the range from 0,5% to 13,5%. The weighted mean level of breakdowns composed 3% yearly (declared by producers HDD - 0,88%). Researcher draws the conclusion that declared by producers the period of time to failure in format on the average, millions of hours, do not correspond to reality. Actually this index composes 300 thousand hours.
Shredder noted that in the case of hdd failure one of disks massif substantially rises the probability of failure for all data , which duplicates one and the same information.
Unfortunately, data according to the producers HDD in this report, as in a study Of google, it was not represented.