US 20050102467 A1
Optimizing performance of physical disk drives in a disk array storage device with a plurality of logical volumes is accomplished by accumulating statistics over an interval to select two logical volumes for an exchange. After testing to determine any adverse effect of making that change, the exchange occurs to more evenly distribute the loading on individual physical disk storage devices. One set of statistics, seek time, is obtained by combining estimated disk access information and an interpolated seek time obtained from a template that stores characteristics seek times among different pairs of segments that define a physical disk device independently of a logical volume configuration.
1. A program for determining the seek time over a time interval for a physical disk drive configured to store data in a plurality of logical volumes, said program comprising:
A) a template procedure for dividing the physical disk into a plurality of fixed sized segments independently of the logical volume configuration on the physical disk drive wherein each segment has a plurality of contiguous tracks,
B) a seek time procedure for determining actual seek times for seek operations between the segments including,
i) a store for assigning a predetermined seek time for each seek operation between two segment boundaries,
ii) a calculator for calculating an intrasegment seek time based upon the predetermined seek times, and
iii) an array generator for placing the predetermined seek times in a two-dimensional array with the rows and columns defined by the segment boundaries.
C) an accumulation procedure for accumulating statistics for each access to each logical volume during the time interval,
D) a conversion procedure for converting the accumulated statistics into an estimated number of seeks between each pair of logical volumes,
E) a seek time generator for defining a seek time for each logical volume pair based upon said actual segment seek times, and
F) a total seek time generator for generating a total seek time as the sum of the products, for each logical volume pair, of the actual seek time for and the estimated number of seeks between each logical volume in the logical volume pair over the time interval.
2. A program as recited in
3. A program as recited in
4. A program as recited in
5. A program as recited in
6. A program as recited in
7. A program as recited in
8. A program as recited in
9. A program as recited in
10. A program as recited in
where x and y represent center line locations of logical volumes, └x┐ and ┌y┐ represent “floor of x” and “ceiling of y” functions for the values of x and y based upon the boundaries of the segments and p and q represent the displacements of the center line addresses for each logical volume relative to a segment boundary; and said seek time generation procedure generates a seek time according to:
where tij represents the seek time for a specific pair of logical volumes and time is the seek interval for the corresponding relationship.
11. Apparatus for determining the seek time over a time interval for a physical disk drive configured to store data in a plurality of logical volumes, said apparatus comprising:
A) dividing means for dividing the physical disk into a plurality of fixed sized segments independently of the logical volume configuration on the physical disk drive wherein each segment has a plurality of contiguous tracks,
B) determination means for determining actual seek times for seek operations between the segments including:
i) means for assigning a predetermined seek time for each seek operation between two segment boundaries,
ii) means for calculating an intrasegment seek time based upon the predetermined seek times, and
iii) means for placing the predetermined seek times in a two-dimensional array with the rows and columns defined by the segment boundaries,
C) accumulating means for accumulating statistics for each access to each logical volume during the time interval,
D) conversion means for converting the accumulated statistics into an estimated number of seeks between each pair of logical volumes,
E) seek time definition means for defining a seek time for each logical volume pair based upon said actual segment seek times, and
F) total seek time generation means for generating a total seek time as the sum of the products, for each logical volume pair, of the actual seek time for and the estimated number of seeks between each logical volume in the logical volume pair over the time interval.
12. Apparatus as recited in
13. Apparatus as recited in
14. Apparatus as recited in
15. Apparatus as recited in
16. Apparatus as recited in
17. Apparatus as recited in
18. Apparatus as recited in
19. Apparatus as recited in
20. Apparatus as recited in
where x and y represent center line locations of logical volumes, └x┘ and ┌y┐ represent “floor of x” and “ceiling of y” functions for the values of x and y based upon the boundaries of the segments and p and q represent the displacements of the center line addresses for each logical volume relative to a segment boundary and wherein said means for defining a seek time for each logical volume includes means for generating a seek time according to:
where tij represents the seek time for a specific pair of logical volumes and time is the seek interval for the corresponding relationship.
This is a continuation of copending application for U.S. patent Ser. No. 09/541,159 filed Mar. 31, 2000 that is a continuation-in-part of copending application for U.S. Pat. No. 6,088,766 granted Jul. 11, 2000 (Ser. No. 09/002,428 filed Jan. 2, 1998) that is a continuation-in-part of copending application for U.S. Pat. No. 6,061,761 granted May 2, 2000 (Ser. No. 08/944,606 filed Oct. 6, 1997) and that is assigned to the same assignee as this invention.
1. Field of the Invention
This invention generally relates to the management of resources in a data processing system and more particularly to a tool for use in the management of a disk array storage device.
2. Description of Related Art
Many data processing systems now incorporate disk array storage devices. Each of these devices comprises a plurality of physical disks arranged into logical volumes. Data on these devices is accessible through various control input/output programs in response to commands, particularly reading and writing commands from one or more host processors. A Symmetrix 5500 series integrated cached disk array that is commercially-available from the assignee of this invention is one example of such a disk array storage device. This particular array comprises multiple physical disk storage devices or physical disk drives with the capability of storing terabytes of data. The management of such resources becomes very important because the ineffective utilization of the capabilities of such an array can affect overall data processing system performance significantly.
Generally a system administrator will, upon initialization of such a direct access storage device, determine certain characteristics of the data sets to be stored. These characteristics include the data set size, and volume names and, in some systems, the correspondence between a logical volume and a particular host processor in a multiple host processor system. The system administrator uses this information to configure the disk array storage device by distributing various data sets across different physical disk devices accordingly with an expectation of avoiding concurrent use of a physical device by multiple applications. Often times allocations based upon this limited information are or become inappropriate. When this occurs, the original configuration can degrade overall data processing system performance dramatically.
One approach to overcoming this problem involves an analysis of the operation of the disk array storage device prior to loading a particular data set and then determining an appropriate location for that data set. For example, U.S. Pat. No. 4,633,387 to Hartung et al. discloses load balancing in a multi-unit data processing system in which a host operates with multiple disk storage units through plural storage directors. In accordance with this approach a least busy storage director requests work to be done from a busier storage director. The busier storage director, as a work sending unit, supplies work to the work requesting, or least busy, storage director.
U.S. Pat. No. 5,239,649 to McBride et al. discloses a system for balancing the load on channel paths during long running applications. In accordance with the load balancing scheme, a selection of volumes is first made from those having an affinity to the calling host. The load across the respective connected channel paths is also calculated. The calculation is weighted to account for different magnitudes of load resulting from different applications and to prefer the selection of volumes connected to the fewest unused channel paths. An optimal volume is selected as the next volume to be processed. The monitored load on each channel path is then updated to include the load associated with the newly selected volume, assuming that the load associated with processing the volume is distributed evenly across the respective connected channel paths. The selection of the following volume is then based on the updated load information. The method continues quickly during subsequent selection of the remaining volumes for processing.
In still another approach, U.S. Pat. No. 3,702,006 to Page discloses load balancing in a data processing system capable of multi-tasking. A count is made of the number of times each I/O device is accessed by each task over a time interval between successive allocation routines. During each allocation, an analysis uses the count and time interval to estimate the utilization of each device due to the current tasks. An estimate is also made with the anticipated utilization due to the task undergoing allocation. The estimated current and anticipated utilization serve as a basis for the allocation of data sets to the least utilized I/O devices.
Yet another load balancing approach involves a division of reading operations among different physical disk drives that are redundant. Redundancy has become a major factor in the implementation of various storage systems and must also be considered in configuring a storage system. U.S. Pat. No. 5,819,310 to Vishlitzky, et al. discloses such a redundant storage system with a disclosed disk array storage device that includes two device controllers and related physical disk drives for storing mirrored data. Each of the physical disk drives is divided into logical volumes. Each device controller can effect different reading processes and includes a correspondence table that establishes the reading process to be used in retrieving data from the corresponding physical disk drive. Each disk controller responds to a read command that identifies a logical volume by using the correspondence table to select the appropriate reading process and by transferring data from the appropriate physical disk drive containing the designated logical volume.
Consequently, when this mirroring system is implemented, reading operations involving a single logical volume do not necessarily occur from a single physical disk drive. Rather read commands to different portions of a particular logical volume may be directed to any one of the mirrors for reading from preselected tracks in the logical volume. Allowing such operations can provide limited load balancing and can reduce seek times.
Other redundancy and striping techniques may spread the load over multiple physical drives by dividing a logical volume into sub-volumes that are stored on individual physical disk drives in blocks of contiguous storage locations. However, if the physical disk drives have multiple logical volumes, sub-volumes or other forms of blocks of contiguous storage locations, the net effect may not balance the load with respect to the totality of the physical disk drives. Thus, none of the foregoing references discloses or suggests a method for providing a dynamic reallocation of physical address space based upon actual usage.
Recently more rigorous analyses have been implemented to provide dynamic reallocation based upon actual usage. U.S. Pat. No. 6,189,071(application Ser. No. 09/143,683 filed, Aug. 28, 1998) discloses one such analysis that includes the step of providing an approximation of disk seek times. Generally these approaches determine seek distances and convert the seek distances into time. In more specific terms, this approach uses a statistical analysis by which actual disk accesses are weighted and combined to produce an estimated seek activity. Then this estimate is converted to a seek time by combination with a value, ti,j, that is an approximation of the seek time between two logical volumes i and j. However, in some applications it may be desirable to obtain more accurate seek times to use in selecting exchangeable logical volumes that, in turn, can optimize the performance of a disk array storage device.
Therefore it is an object of this invention to provide a tool that is useful in the dynamic reallocation of data on a disk array storage device based upon actual usage.
Another object of this invention is to provide dynamic reallocation of a disk array storage device utilizing statistical analyses based, in part, on better approximations of disk seek times.
Still another object of this invention is to provide a dynamic reallocation of a disk array storage device by providing a method for generating disk seek times directly from normally collected disk statistics based upon physical disk drive operating characteristics.
Yet another object of this invention is to provide a statistical analysis that includes a process for generating disk seek times that more accurately reflect actual disk seek times thereby to facilitate the optimization of performance.
In accordance with this invention, the seek time required for moving a disk head between first and second addresses includes a step of dividing the disk into a plurality of segments of a given size defined by first and second boundaries and establishing an array of actual seek times for seek operations between each pair of segments. The specified first and second addresses and actual seek times in the array are combined to obtain an actual seek time for a seek between the first and second addresses.
In accordance with another aspect of this invention the seek time for a physical disk drive configured to store data in a plurality of logical volumes over a time interval is determined by dividing the physical disk into a plurality of fixed sized segments. Segment size is independent of the logical volume configuration on the physical disk drive. The actual time required for seek operations between the segments is obtained. Access statistics for each logical volume during the time interval are accumulated and converted into an estimated number of seeks between each pair of logical volumes. A characteristic seek time for each logical volume pair is obtained based upon actual segment seek times. Then a total seek time is generated as the sum of the products, for each logical volume pair, of the characteristic seek time and the estimated number of seeks for the logical volume pair.
The appended claims particularly point out and distinctly claim the subject matter of this invention. The various objects, advantages and novel features of this invention will be more fully apparent from a reading of the following detailed description in conjunction with the accompanying drawings in which like reference numerals refer to like parts, and in which:
Configuring physical disk drives into logical volumes or analogous data blocks is well known in the art. By way of example, the device controller 34A is shown with an associated physical disk drive 31A divided into the mirrored logical volumes M1-LVA, M1-LVB, M1-LVC and M1-LVD; the device controller 34E controls the other physical disk drive 31E that stores the mirrored logical volumes M2-LVA, M2-LVB, M2-LVC and M2-LVD. It is also possible for logical volumes, such as the LVE and LVF logical volumes on the physical disk 31B, and LVG and LVH logical volumes in the physical disk drive 31C to have different capacities. As will become apparent, this invention has particular use when a physical disk drive has multiple logical volumes.
Still referring to
A system manager console 40 includes an additional processor that connects to the system bus 38, typically through a device controller, such as device controller 34A by means of a serial or other communications link. Essentially the system manager console 40 enables the operator to establish communications with the host adapter 37, the device controller 34B and the system memory 32. As known, a system manager 47 in the system manager console 40 permits a system operator to run set-up and diagnostic programs for configuring, controlling and monitoring the performance of the data storage facility 30.
Before any component, such as the host adapter 37 or the device controllers 34A and 34B can access the system memory 32, that component must obtain access to the system bus 38. Conventional bus access logic 41 receives access request signals from these components and grants access to only one such component at any given time. A wide variety of known arbitration schemes are suitable for use in a data storage system employing multiple processors and a shared system memory, such as the system memory 32.
Preferably the system memory 32 is a high-speed random-access semiconductor memory that includes, as additional components, a cache index directory 42 that provides an indication including the addresses of the data which is stored in the cache memory 33. In a preferred embodiment, the cache index directory 42 is organized as a hierarchy of tables that define a logical device or volume as a group of cylinders, individual cylinders within a logical volume, and individual tracks with a cylinder. The system memory 32 also includes areas for path structures 43 and queues 44. The basic operation of the system memory 32 is described in Yanai et al., U.S. Pat. No. 5,206,939 issued Apr. 27, 1993. System memory 32, particularly the cache memory 33, may also include a region of memory known as permacache memory that retains data elements until they are specifically deleted.
The coordination of each of the host adapters with each of the device controllers is simplified by using the system memory 32, and in particular the cache memory 33, as a buffer for data transfers between each host adapter and each device controller. Such a system, for example, is described in U.S. Pat. No. 5,206,939.
A cache memory manager 45 in the host adapter 37 gathers statistics about a number of parameters. For the purpose of this invention, the numbers of reading and writing operations requested by a host 35 or connected hosts are important. Each of the cache memory managers 46A through 46X in each of the device controllers 34A through 34X gathers statistics for the logical volumes on each connected physical disk storage device. More specifically, each cache memory manager gathers statistics as to the number of disk accesses to each logical volume connected to it. A monitor 50 in the system manager console 40 integrates the data in these cache memory managers to obtain appropriate statistics at given intervals.
From the foregoing, disk operations included in any measure of the loading of a logical volume will include reading operations and writing operations. Reading operations can be further segregated or classified as read-hit, read-miss and sequential read operations. A read-hit operation occurs when the data to be read resides in the cache memory 33. A read-miss occurs when the data to be read is not available in the cache memory 33 and must be transferred from a physical disk device. Sequential read operations are those that occur from sequentially addressed storage locations.
The system operates with two types of writing operations. The first transfers the data from the host 35 to the cache memory 33. The second type transfers the data from the cache memory 33 to a physical disk storage device. The second type operates in a background mode, so it is possible that the host 35 may write data to a location more than once before the data is written to a physical disk storage device. Consequently the number of writing operations of the second type normally will not correspond to and be less than the number of writing operations of the first type.
In accordance with this invention, optimization of disk performance relies upon the foregoing information supplied from a performance monitor 50 that retrieves statistics from each cache memory manager on a periodic basis. The periodicity will be selected according to conventional sampling criteria. Typical periods will be from up to 15 to 30 or more minutes. As each set of statistics is time stamped and accumulated by logical volume, the total number of read operations, a read-hit ratio, a sequential-read ratio and the total number of writing operations over a test interval can be obtained. An optimization program 51 shown in
Step 111 in
The procedure of step 114 converts the weighted disk activity into a disk transfer time representing the time to complete all the transfers to the logical volume exclusive of any seek operations. That is, a disk transfer time will correspond to any latency time plus the time for actually transferring selected data. This conversion can be provided by data contained in a table that represents an average of all systems or specific systems by model and manufacturer. The data may be manufacturer's design data or may reflect specific measurements at one track on a physical disk drive or at a plurality of spaced tracks.
Once this information has been calculated for a particular logical volume or other data block, step 115 determines whether additional logical volumes remain to be tested. If more logical volumes exist, steps 112 through 114 repeat.
After all the logical volumes in all the physical disk drives under analysis have been processed to obtain the disk transfer times, step 115 diverts control to step 116 to begin an analysis that provides seek times for the accesses. Specifically step 116 selects a physical drive for an analysis according to a procedure 117, shown in detail in
In accordance with this invention the procedure for determining the seek time for a physical disk drive in step 117 is based upon an assumption that each physical disk drive is divided into segments of a given size defined by segment boundaries, notwithstanding the boundaries of any logical volumes on the physical disk drive. For each physical disk drive the segment boundaries are utilized to form a template that is common to all physical disk drives of the same type, notwithstanding the number or sizes of the logical volumes on a particular physical disk drive.
This invention utilizes a template that increases the accuracy of the estimates of seek time during an analysis.
Step 207 segments the drive into blocks of a given size with identified boundaries. The actual selection of the segment boundaries will depend upon the data that is available for determining seek times. For this particular example, it is assumed that step 207 segments the physical disk drive 200 into 1 GB segments with boundaries at 0, 1, 2, 3 and 4 GB as defined above. As will be apparent from
Step 208 prepares a template with empirical predetermined seek times based upon the identified boundaries.
Step 210 stores the seek time template 209 for the device type. In one embodiment the template is stored within data buffer space provided within the load balance system 51. Obviously this template could be stored in any accessible storage location in the data storage facility 30 shown in
This invention uses the foregoing seek time template in
An intervolume seek procedure 220 in
As will be apparent from
Assume that the LV(1)-LV(3) logical volume pair is selected for analysis, as disclosed in the above table and
Step 226 utilizes the resulting number of seeks obtained from Equation (2) and the characteristic seek time from equation (9) to generate the total time for seek operations with the logical volume pair. That is, for a specified logical volume pair i,j, the seek time, seek time(i,j) is:
There are two specific intervolume seek operations in which individual terms of equation (9) can not be solved using the data in a seek time template as shown in
Step 226 uses one of this set of three equations to provide an appropriate value for the interpolation of the seek time. In the specific example shown in
Step 231 obtains appropriate data from the system manager 40 in
Step 234 determines the boundary locations on the physical disk drive for the selected logical volume. For the specific instance of the LV(2) logical volume, the boundaries are at 1.7 GB and 2.7 GB. Step 235 uses this information to determine the estimated seek time according to:
For intravolume seek operations with the selection of the boundaries, tseek(i,i) is taken between the starting and ending locations for the logical volume. However, the exact extent of any seek operation may be from a minimum between two adjacent cylinders to a maximum between the starting and ending locations. Information obtained from experimentation indicates that a value K=⅓ scales the total seek time to a more accurate value. Thus, equation (16) provides an accurate estimate of the seek time for seek operations that are confined to a specific logical volume.
As will become evident, it is only necessary to analyze the intervolume seek times for each logical volume pair. Any number of procedures can be used to avoid duplication. In one approach depicted in
Step 241 defines an LViPTR pointer and initializes it to point to the first logical volume in the ordered sequence. In the example of
Step 242 selects the logical volume i from the available logical volumes. Step 243 then applies the procedure 230 in
Step 245 then initializes an LVjPTR pointer to point to the next logical volume in the ordered sequence. For a first iteration, the LViPTR pointer points to the LV1 logical volume in
Step 251 tests to see if the LVjPPT pointer identifies the last logical volume in the ordered sequence. If not, step 252 increments the LVjPTR to point to a next logical volume for being paired with the logical volume identified by the LViPTR pointer and transfers to step 246 to select that logical volume and the statistics for the logical volume pair; e.g., for the LV1-LV3 logical volume pair.
After evaluation all the possible logical volume pairs in which the first logical volume in the ordered sequence is a member, step 251 transfers control to step 253 determines whether the LViPTR pointer identifies the last logical volume in the ordered sequence. If not, step 254 increments the LViPTR pointer. This establishes a reference to the LV2 logical volume after all the possible LV1 logical volume pairs are evaluated and control transfers to step 242 to select the next logical volume and its seek statistics.
For the specific example of
When all logical volumes are tested, step 253 transfers control to step 255 that records the value in the seek time register in a memory location assigned to the selected physical disk drive. That is, after step 255 this assigned memory location will contain a value for the total seek time for the physical disk drive that is given by:
As will now be appreciated, using fixed size segments allows the generation of a seek time template that is independent of the logical volume configuration on a particular physical disk drive of a given drive type. Many systems that comprise multiple disk drives include physical disk drives of a single type in which case only one such template is required. The 4 GB physical disk drive with 1 GB segments has been selected because it simplifies the description of the invention. In most applications the physical disk drive will have a greater storage capacity and may include any number of logical volumes. The segment sizes may also differ.
The process of obtaining seek times in accordance with this invention occurs with only a minimal impact on processing within the system. That is, reading the size of each logical volume, calculating its center line and performing the interpolation as set forth in Equations (3) through (10) only requires minimal processing time but improves the analysis of the seek times.
In summary, this foregoing disclosure defines a method for determining seek times that is useful in optimizing the performance of a disk storage facility comprising a plurality of physical disk drives. Statistics of the occurrence of read, write, and sequential prefetch read operations are maintained over at least an analysis interval. This information becomes an estimated access number for each logical volume. In turn this information produces an estimated number of seeks between individual logical volumes in logical volume pairs. Using a center location for each logical volume as a reference and a drive seek time template enables a rapid, straightforward and more accurate determination of the seek time for each logical volume pair. These times are easily combined to produce a total seek time for the analysis interval for the physical disk drive.
The foregoing description discusses this invention in terms of data organized into blocks of contiguous storage locations on a physical disk of known size called logical volumes. However, the invention is applicable to other data organizations. In some applications, for example, a logical volume might be divided into a series of sub-volumes distributed across plural physical disk storage devices or other logical blocks. Such a division could be made for redundancy and recovery purposes or for load distribution purposes.
Although this invention has been disclosed in terms of certain embodiments, many modifications can be made to the disclosed method and supporting data organization without departing from the invention. Therefore, it is the intent of the appended claims to cover all such variations and modifications as come within the true spirit and scope of this invention.