Disk arrays are widely used, but they are often the bottleneck for the secure operation of the entire system. The reliability of quantitative computing is getting more and more attention, because of the quantitative data, it can provide a scientific basis for in-depth understanding of the nature of things, lay the foundation for multi-solution optimization and comparison, and then make a correct decision. It is not difficult to make a feasible disk array solution, but it is difficult to make an optimization solution. We believe that the key to solving these problems is whether a mathematical model of reliability can be found. If so, the problem can be solved. In this case, due to actual work needs, I have made some data collection and sorting, and made an exploratory solution to the reliability of the hard disk array. Now I am writing this article to give a reference.
Raid Overview
Raid was proposed by Professor DA Patterson at UC Berkeley. Raid is the abbreviation of Redundant Array of Inexpensive Disks, referred to as "disk array ". Raid can be understood as an organic combination of a group of disk drives to form a Logical Disk Drive. The specific implementation of RAID can rely on hardware, such as disk array cabinets, or software. The windows NT operating system provides the software raid function. Previously, raid was generally implemented using a SCSI disk drive. As the size and quality of the IDE Hard Drive have been significantly improved, raid is now applied to the IDE disk drive. However, because common ide channels can only connect up to four disk drives, the raid functions of common IDE are not as rich as that of SCSI. Now there is a dedicated array cabinet for IDE hard disks, and it has the RAID 5 function. It supports mounting 6 hard disks. At the same time, the emergence of optical interfaces also enriches raid products. From a mathematical point of view, all except raid0 are work redundancy systems.
Raid features
1. low cost, low power consumption, and high data transmission rate. In raid, many disk drives can transmit data at the same time, and these disk drives are logically the same disk drive. Therefore, using raid can speed up to several times of a single disk drive.
2. Fault tolerance can be provided to improve reliability. Of course, this is at the cost of redundancy. This is one of the important reasons why raid is widely used.
3. Raid is a cost-effective and easy way to obtain large-capacity logical memory. This is another important reason why raid is widely used.
4. When RAID is implemented by the hard disk array Cabinet, the price of the hard disk array cabinet is usually more expensive than or even more expensive than that of the hard disk.
Classification and features of commonly used RAID
1. RAID0: No redundant and non-verified disk arrays. Data is distributed across various disk drives at the same time, with no fault tolerance capability, the fastest read/write speed, and the worst reliability. In mathematics, RAID0 is a series model, and the number of disks N can be greater than 2. In this paper, only the mathematical model and reliability of N = 2 are given.
2. RAID1: Image disk array. The most basic parallel model is the 1/2 table Terminator.
3. RAID 0 + 1 is actually derived from RAID 0 and RAID 1. RAID 0 and RAID 1 are performed first. The most basic series-parallel model.
4. RAID 1 + 0 is also derived from RAID 1 and RAID 0. RAID 1 first and RAID 0 again. The most basic series model. The number of series N can be greater than 2.
5. RAID2 ~ 4. This document does not describe or analyze it.
6. RAID5: No parity disk array for independent disk verification. The verification information is distributed on each disk drive. RAID5 has good performance in reading and writing large amounts of data. It can obtain high reliability and make up a large capacity. It also has a flexible choice for the number of disks, therefore, it is widely used. RAID5 is a mathematical structure that belongs to the (N-1)/N table deciders.
7. There are other types of RAID. Although some types and names are the same, each manufacturer has their own special definitions, which should be noted during use.
RAID Reliability Model
The model mentioned here refers to the reliability model, which cannot be confused with the series-parallel connection on the road. For example, if two capacitors are connected in parallel but one fails, the system will be invalidated. In reliability calculation, the two capacitors are considered as series. The reliability model also has a non-series parallel model. The Reliability mentioned here refers to the ability to complete the specified function within the specified time and under the specified conditions. This article calculates the probability.
Disk Array Reliability Calculation
Based on the average failure interval (MTBF), we can infer the failure-free reliability probability of a single disk at the specified time. Based on the data provided by Seagate, the MTBF of the SCSI hard disk is 1 million hours, And the MTBF of the IDE hard disk is 0.4 million hours. MTBF is not provided for hard disks of some brands. It should be said that Seagate's materials are representative. Calculate the reliability of common RAID based on the mathematical model of RAID reliability, and sort the data into tables for comparison. It is not difficult to see from this table that raid 1 of IDE also has a high reliability. Because SCSI hard disks are much more expensive and have higher performance than IDE hard disks, you can choose the following methods to compare them: within the specified time, at the same price; or the price under the same reliability within the specified time; of course there are other comparison methods. It is not difficult to draw the following conclusion: the one-time investment of SCSI hard disks is large, and RAID is more available than once and for all. The one-time investment of IDE hard disks is small. To achieve the required reliability, you must invest in segments. The total price is still low, which is relatively troublesome. Currently, there are few options available, but the increase of options is an inevitable trend. As computer hardware develops rapidly, the above conclusion is only characteristic. The specific process depends on the specific situation.
Disk Array reliability table
One year, two years, and three years
Format/disk SCSI hard drive fault-free reliability (MTBF = 1 million hours)
Single hard drive 0.991278257 0.982632582 0.974062314
Raid 0 0.982632583 0.965566792 0.948797391
Raid 0 + 1 0.999698372 0.998814354 0.997378292
Raid 1 + 0 0.999847868 0.999396836 0.998654925
Raid 1 0.999923931 0.999698372 0.999327236
Raid 5 (2/3) 0.99977312 0.999105595 0.998016609
Raid 5 (3/4) 0.999548877 0.998231871 0.99610166
Raid 5 (4/5) 0.999252494 0.997087139 0.993614621
Raid 5 (5/6) 0.998885246 0.995681075 0.990586456
Raid 5 (6/7) 0.998448393 0.994023102 0.987046811
Format/disk IDE hard drive fault-free reliability (MTBF = 0.4 million hours)
Single hard drive 0.978338064 0.957145367 0.936411745
Raid 0 0.957145367 0.916127254 0.876866957
Raid 0 + 1 0.99816348 0.992965362 0.984838253
Raid 1 + 0 0.999061741 0.996330333 0.991929417
Raid 1 0.99953076 0.99816348 0.995956533
Raid 5 (2/3) 0.9986126109 0.9946478482 0.9883838357
Raid 5 (3/4) 0.9972652197 0.9896003914 0.9777470902
Raid 5 (4/5) 0.9955076143 0.9831588582 0.9644665925
Raid 5 (5/6) 0.9933581989 0.9754520036 0.9489215748
Conclusion:
The MTBF provided by the hard disk producer is a conclusion made under specific conditions. Generally, the actual use environment is different and generally does not meet the conditions described by the manufacturer. Hard disks used by the manufacturer for testing generally do not require long-distance transportation, and the hard disks used by the customer are products that have had transportation experience, or even bad shipment. In addition, the usage skills cannot be ignored, we hope that you will notice these issues. Therefore, the actual reliability will decrease, but the reliability relationship between various forms of RAID exists objectively.
