Electrical computers and digital data processing systems: input/ – Input/output data processing – Input/output command process
Reexamination Certificate
2000-02-02
2002-11-12
Shin, Christopher B. (Department: 2182)
Electrical computers and digital data processing systems: input/
Input/output data processing
Input/output command process
C710S045000, C709S241000, C711S114000, C711S151000
Reexamination Certificate
active
06480904
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to disk time-sharing apparatus and method for scheduling the use of a disk apparatus on the basis of a plurality of inputs and outputs and, more particularly, to disk time-sharing apparatus and method for scheduling the use of a disk apparatus so as to sequentially switch allocating time for inputs and outputs which compete.
2. Description of the Related Arts
Hitherto, in a storage system for managing data by using a disk apparatus such as a hard disk drive or the like, for example, the disk apparatus is constructed so as to have an RAID structure, the RAID apparatus is connected subordinate to a disk control apparatus, thereby processing an input/output from an upper host, or the RAID apparatus is directly connected to a server, thereby processing an input/output from a server OS. In such a storage system, in the case where it is necessary that a random access in which a guarantee of a response time is required and a sequential access in which importance is attached to an amount of processes per unit time are performed for the same disk apparatus, the operation is performed in a time-sharing manner lest the random access and the sequential access compete. For example, an OLTP (On Line Transaction Processing) in which the random access is mainly performed is executed to a database of the disk apparatus in the daytime and a backup of the database is performed at night after the processing.
(Resource distribution of random access and sequential access)
In the storage system, however, in association with the realization of a non-stop processing, the OLTP processing of the random access system needs to be continued even at night, so that it is necessary to execute the backup as a sequential access during the OLTP processing of the random access system. In case of only the random access, an IOPS (Input Output Per Second) such as 100 IOPS as the number of inputting/outputting times per unit time which can satisfy a mean certain response time, for example, 30 milliseconds can be estimated. In case of only the sequential access, a throughput such as 20 MB/sec can be estimated. However, when the random access and the sequential access are simultaneously performed, since received input/output requests are processed by a queue using an FIFO, there is no mechanism to guarantee a period of time during which the random access can use the disk apparatus and a period of time during which the sequential access can use the disk apparatus. For instance, even when the random access of 50 IOPS at mean response time of 30 milliseconds and the sequential access of 5 MB/sec are desired, if the sequential access frequently occurs, the throughput of the sequential access rises from 5 MB/sec to 10 MB/sec although it doesn't need to rise. On the contrary, the IOPS which satisfies the mean response time of 30 milliseconds in the random access deteriorates from 50 IOPS to 25 IOPS although the user doesn't want to reduce it.
(Resource distribution between logic volumes)
In the conventional storage system, by arranging data having different performance requirements to the different disk apparatuses, their performance characteristics are drawn out. For example, data in which a guarantee of a response time is required in the random access of a small amount of data and data in which importance is attached to a processing amount per unit time in the sequential access of a large amount of data are arranged in the different disk apparatuses. In association with the realization of a large capacity of the disk apparatus, however, the case where the data having different performance requirements is arranged in the same disk apparatus is increasing. A similar problem occurs even when the logic volumes having different performance requirements are arranged to the same disk as mentioned above. Hitherto, there is not a mechanism for controlling a disk resource distribution between logic volumes by scheduling the received inputs/outputs by the FIFO. Therefore, when the input/output to/from a certain logic volume frequently occurs, input/output performance for the other logic volume deteriorates. For instance, in the case where a volume A in which it is desired to guarantee 10 IOPS and a volume B in which it is desired to guarantee 50 IOPS are arranged on the same disk, when the access to the volume A frequently occurs, the IOPS of the volume A rises from 10 IOPS to 20 IOPS though it doesn't need to rise. On the contrary, the IOPS of the volume B deteriorates from 50 IOPS to 40 IOPS though it is not desired to deteriorate it.
(Resource distribution between normal process and backup/copying process)
A case where a plurality of logic volumes exist on the same disk apparatus in the conventional storage system and a backup or a copying operation is performed on each logic volume unit basis will now be considered. Hitherto, in order to suppress the influence on the normal input/output by the backup/copying process, a method of setting paces (intervals) of the backup/copying process at the time of executing the backup/copying process is used. However, if the copying operation is executed to the volume B on the same disk apparatus as that of the volume A while the volume A is being copied, the duplex copying process is operated simultaneously on the same disk apparatus, so that the influence on the normal input/output is doubled.
(Resource distribution between normal process and rebuilding)
In the RAID apparatus, by making data redundant in a plurality of disk drives, even if a failure occurs in one disk drive, the data can be recovered from the remaining disk drives. In the RAID apparatus, therefore, even if the failure occurs in the disk drive, the ordinary input/output can be continued. A recovery of the data is performed to the exchanged disk drive from the remaining disk drives. The recovering process is called “rebuilding”. Since the rebuilding is accompanied with the input/output process for the disk drives constructing the RAID apparatus, the rebuilding and the normal input/output scramble for the same disk drive. Consequently, the performance of the normal input/output is deteriorated by the rebuilding. For example, in case of RAID
1
having a mirror construction, the rebuilding is a process for copying data from one disk drive which remains due to the failure of the other disk drive to the exchanged new disk drive and a read input/output occurs to the disk drive on the copying source side. The read input/output causes the normal input/output to wait, so that the performance of the normal input/output deteriorates. There are two conventional approaches to solve the problem. According to the first approach, enough small data is copied at an enough long interval so as not to exert an influence on the normal input/output. In this case, although the influence on the normal input/output can be reduced, time that is required until the rebuilding is completed becomes long. For instance, in case of RAID
1
constructed by disk drives of 9 GB, time of about 10 hours is needed. As for the second approach, when the disk drive is vacant, namely, when the disk drive is not used in the normal input/output, the input/output of the rebuilding is scheduled. A problem in this case is a point that the time that is required until the completion of the rebuilding cannot be guaranteed. When the disk drive is hardly vacant, long time is needed for the rebuilding.
(Guarantee of maximum response time)
In a mission critical processing, as requirements of the input/output performance, the maximum response time is important in addition to the mean response time. The recent disk apparatus has a re-ordering function for rearranging inputs/outputs for which the execution is waited so as to minimize the processing time. The re-ordering function is a function such that an input/output to minimize a positioning time that is defined by the sum of a seeking time and a rotation waiting time is selected as an input/output to be subsequently executed from the ex
Hotta Yuuji
Ito Mikio
Kato Tadaomi
Kimura Osamu
Take Riichiro
LandOfFree
Disk-time-sharing apparatus and method does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Disk-time-sharing apparatus and method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Disk-time-sharing apparatus and method will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2923425