Load balancing in a multiunit system

• US 4,633,387 A
• Filed: 02/25/1983
• Issued: 12/30/1986
• Est. Priority Date: 02/25/1983
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Reed, David G., Tayler, Gerald E., Nolta, Arthur H., Hartung, Michael H.
• Primary Examiner(s): Thomas, James D.
• Assistant Examiner(s): Lee, Thomas C.
• Application Number:
  US06/469,680
• Time in Patent Office: 1,404 Days
• Field of Search: 364/200 MS File, 364/ 900 MS File, 364/ 300
• US Class Current: 718/105
• CPC Class Codes: G06F 12/0866 for peripheral storage systems, e.g. disk cache
• CPC Class Codes: G06F 2212/312 In storage controller

Abstract

In a multiunit data processing system, such as a multicontrol unit peripheral data storage system, a least busy one of the units requests work to be done from a busier unit. The busier unit, a work sending unit, supplies work to the work requesting or least busy unit. Work thresholds in the respective units determine when work is to be requested or transferred. In a data storage environment, the transferred work consists of data transfers to be achieved usually asynchronously to connected host activities, such as data transfers between a backing and a front store in a data storage hierarchy.

Claims

1. A load-balancing control apparatus for balancing asynchronous activity in a multiunit system, including transferring data signals between any one of a first plurality of first units and a second unit via any one of a second plurality of transfer path means, comprising:

means for maintaining a queue of work to be done that relates to said data transfers for each of said transfer path means;

means responsive to each said path means, respectively, for sensing the level of work to be done within said queue, determining the volume of work for said path means, and indicating a low volume of work for said path means;

requesting means responsive to each of said path means and with said low-volume indicating means for sending a proxy request message which requests the transferring of a proxy work job related to said data transfers to a second of said path means from a first of said path means;

responding means in said second path means coupled to said requesting means for supplying the proxy work job related to said data transfer thereto; and

operation control means in said first path means and coupled to said responding means, to said first units and to said second unit for performing said proxy work job received from said second path means and further having means for performing such operations independently of said proxy work job to effectively balance workload among the plurality of transfer path means in said multiunit system.

2. In a load-balancing control apparatus as set forth in claim 1 wherein each of said path means is a storage director, and said first units are data storage devices (DASD) and said second unit is a data storage cache;

each of said storage directors including means for receiving requests from an external source for accessing said DASD and having means for servicing such received requests including means for interleaving operations relating to said asynchronous activity with operations synchronous to the operation of said external source further comprising;

means in the storage director which is said second path means for indicating work status thereof and being coupled to said responding means for indicating thereto the level of work in said second path means; and

means in said responding means for receiving said work status indication for preventing transferring said proxy work job when said work status indication indicates a predetermined low work level whereby load-balancing by proxy work transfers are limited to storage directors having a predetermined work level.

3. A load-balancing control apparatus as set forth in claim 2 wherein the queue maintaining means further includes a read queue in each of said storage directors for indicating those asynchronous activity items relating to transferring data from said DASD to said cache and a write queue in each of said storage directors for indicating those asynchronous activity items relating to transferring data from said cache to said DASD;

and

means in each of said storage directors coupled to said queues being a part of said work status means for indicating the level of work to be done by each of said coupled queues whereby the total work in each respective storage director is indicated via said queues.

4. A load-balancing control apparatus as set forth in claim 2, wherein the queue maintaining means further includes a write queue common to all of said storage directors and a read queue in each of said storage directors which is independent of all other read queues;

means in said requesting means coupled to said common write queue for initiating asynchronous write activity for all of the storage directors directly from said common write queue and said low-volume indicating means operative only with respect to and being coupled to said independent read queues, respectively; and

means in said requesting means coupled to said common write queue means and to sai low-volume means for selecting a priority of operation therebetween.

5. In a machine-implemented method of operating a plurality of programmable units, each programmable unit having a plurality of asynchronous work-performing portions, message means coupling all of said programmable units for enabling communications therebetween, said asynchronous work performing portions performing similar work such that asynchronous work of a first programmable unit can be performed by a second of said programmable units;

the improvement including the steps of;

in each said programmable units maintaining a work queue for scheduling work to be done in the respective asynchronous portion of said programmable units;

sensing said work queue in said first programmable unit for determining whether or not a light load condition prevails;

if no such light load condition prevails, continuing to operate said asynchronous portion independent of said second programmable unit;

upon sensing said relatively light load condition, sending a message to said second programmable unit via said message means requesting that work on said work queue of said second unit be transferred to said first programmable unit;

in said second programmable unit, responding to said message, examining its said work queue and if a given work level exists, sending an identification of predetermined work via said message means to said first programmable unit;

performing said predetermined work in said first programmable unit as if such predetermined work was being done by said second programmable unit; and

upon completion of said predetermined work by said first programmable unit, in said first programmable unit sending a completion message to said second programmable unit and in said second programmable unit post processing said predetermined work as if completed within said second programmable unit.

6. In the machine-implemented method set forth in claim 5 including least recently used means coupled to all of said programmable units for indicating status of a first shared data storage unit used in connection with all of said asynchronous portions;

further including the steps of;

in said first programmable unit, before sending completion message, accessing said least recently used means and updating same to reflect the status of said shared data storage unit after completion of said predetermined work; and

sending said completion message in a first form when a first predetermined work type was performed and in a second form when a second predetermined work type was performed for said predetermined work.

7. The machine-implemented method according to claim 6 wherein said predetermined work of a first type is the step of transferring data signals from said first data storage unit to an addressed one of a plurality of second data storage units, said predetermined work of a second type comprises the step of transferring data signals from said addressed one of said second data storage units to said first data storage unit and

sending said completion message from said first programmable unit for said second type of predetermined work to said second programmable unit in a form for enabling said second programmable unit to continue further processing of such transfer data signals from said first data storage unit and sending said completion message for said first type of predetermined work only for erasing same from its respective work queue.

8. A data processing system having a plurality of intercoupled programmable units having a work queue for queuing and performing a number of predetermined data processing operations such that any one of said programmable units can perform any one of said queued data processing operations, each of said programmable units having a data storage unit for storing electrically sensible indicia representing programs of instructions executable by the respective programmable units by sensing said stored electrical signals, respectively, for performing said queued data processing operations, further comprising:

first program indicia means in the data storage unit of each of said programmable units for enabling each programmable unit to scan for said work queue and initiate execution of said queued data processing operations;

second program indicia means in the data storage unit in each of said programmable units for enabling each programmable unit to measure the level of work to be done within the work queue and indicate its load condition including indicating a light load condition;

third program indicia means in the data storage unit of each of said programmable units for enabling a first programmable unit to generate and send to a second of said programmable units a message requesting a proxy assignment of one of said data processing operations queued in said second programmable unit and whereby such programmable unit sending said message being a light load programmable unit;

fourth program indicia means in the data storage unit of each of said programmable units for enabling each programmable unit to receive and respond as said second programmable unit to a message recieved from said light load programmable unit requesting said proxy assignment of said one queued data processing operation;

fifth program indicia means in the data storage unit of each of said programmable units for enabling each programmable unit as said second programmable unit to select a given one of its queued data processing operations for performance by said light load programmable unit such that said fourth program indicia enablement of such second programmable unit transfers to said light load programmable unit as proxy work such selected queud data processing operation;

sixth program indicia means in he data storage unit of each of said programmable units for enabling each programmable unit to receive only after sending as a light load programmable unit a proxy request message to said second programmable unit any selected queued data processing operation supplied by said second programmable unit as enabled by said fourth program indicia of said second programmable unit;

other program indicia means in said data storage unit of light load programmable unit for enabling said light load programmable unit to perform the data processing operation received from said second programmable unit; and

seventh program indicia means in each of said programmable units for enabling said light load programmable unit to report completion of said selected queued data processing operation to said second programmable unit whereby any of said programmable units can at different times by said second or said light load programmable unit such that queued data processing operations are transferred to said light load programmable unit at the request of such light load programmable unit.