A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.
This application is related to the following patent applications, each of which is hereby incorporated by reference in its entirety:
U.S. Patent Application entitled “SYSTEM AND METHOD FOR USING A SEQUENCER IN A CONCURRENT PRIORITY QUEUE”, application Ser. No. 13/781,493, filed Feb. 28, 2013, by inventor Oleksandr Otenko, and subsequently granted as U.S. Pat. No. 9,110,715 issued on Aug. 18, 2015;
U.S. Patent Application entitled “SYSTEM AND METHOD FOR TRANSFORMING A QUEUE FROM NON-BLOCKING TO BLOCKING”, application Ser. No. 13/781,500, filed Feb. 28, 2013, by inventor Oleksandr Otenko, which is currently pending.
The present invention is generally related to computer systems and software such as middleware, and is particularly related to systems and methods for supporting queue in a middleware machine environment.
Within any large organization, over the span of many years, the organization often finds itself with a sprawling IT infrastructure that encompasses a variety of different computer hardware, operating systems, and application software. Although each individual component of such infrastructure might itself be well-engineered and well-maintained, when attempts are made to interconnect such components, or to share common resources, it is often a difficult administrative task. In recent years, organizations have turned their attention to technologies such as virtualization and centralized storage, and even more recently cloud computing, which can provide the basis for a shared infrastructure. However, there are few all-in-one platforms that are particularly suited for use in such environments. These are the general areas that embodiments of the invention are intended to address.
Systems and methods are provided for supporting cooperative concurrency in a priority queue. The priority queue, which includes a calendar ring and a fast lane, can detect one or more threads that contend to claim one or more requests in the priority queue. Then, a victim thread can place the request in the fast lane in the priority queue, and release a contending thread, which proceeds to consume the request in the fast lane.
Other objects and advantages of the present invention will become apparent to those skilled in the art from the following detailed description of the various embodiments, when read in light of the accompanying drawings.
Described herein are systems and methods that can support cooperative concurrency in a middleware machine environment.
Priority Queue
In accordance with various embodiments of the invention, a concurrent system can use a priority queue to prioritize incoming requests in order to provide service with an appropriate service level agreement (SLA).
The priority queue 301 can be designed to meet demanding concurrency criteria, so that the interaction between the contenders does not cause degradation in the throughput of the system as a whole. Additionally, the priority queue 301 can be implemented to have a fixed memory footprint, so that the JVM is able to better optimize its operations on fixed-size arrays of primitives, and can achieve substantial cache efficiency.
In accordance with various embodiments of the invention, the priority queue 301 can be implemented based on a calendar queue, e.g. the calendar queue provided in the WebLogic Application Server. The calendar queue can include a calendar with multiple buckets, each of which can store events that fall within a particular slice of time. For example, the multiple buckets can be sorted and arranged by comparing the target service time with a current time. If the difference in time is in the first byte, then the request can be stored in a bucket in the first 256 buckets. The specific bucket can be chosen using the actual value of the target time for executing the request. Furthermore, if the difference in time is in the second byte, then the request can be stored in a bucket in the second 256 buckets.
When a consumer, e.g. via one of the worker threads A-C 321-323, tries to remove the next request that is configured to be execute the earliest, the system can scan the calendar for the first bucket that is not empty. If this bucket is not one of the first 256 buckets, then the calendar queue can use a loop and promote method to move the requests to the buckets “one level down” toward the first 256 buckets. Eventually, some requests can be promoted to one or more buckets in the first 256 buckets, and the consumer can claim a request and proceed accordingly.
The above promotion process can involve logarithmic cost, which may have an impact on the overall performance of the system. Additionally, there can be other designs for the calendar queue, the performance of which may be limited to a choice between “O(1) add, O(log N) delete_min,” and “O(log N) add, O(1) delete_min.”
The request manager 402, which manages a thread pool 403, can have a separate logic for associating different threads with different requests. For example, the request manager 402 can serialize all thread pool method calls by wrapping the calls to the priority queue 401 in a synchronized statement, or a synchronized block 410, using a lock mechanism.
Thus, the operations on the priority queue 401 may be limited by the single-threaded design since the serialization is done outside the non-blocking priority queue 401.
Concurrent Priority Queue
The concurrent priority queue 501 can include a calendar, e.g. a calendar ring 502, which is capable of prioritizing and storing incoming requests. The calendar ring 502, the size of which is limited, can be configured to store requests that have a target response time within a preconfigured time limit. Within the calendar ring 502, a request can be stored, or placed, directly in the ring buffer at a position that matches Quality of Service (QoS) of the request, e.g. the target service time.
Thus, the system can achieve a much cheaper lookup for requests without changing the memory footprint of a calendar queue. Furthermore, the system can reduce the logarithmic complexity of the delete_min operation of the calendar queue to mostly a linear cache efficient search, while keeping the adding of elements to the calendar queue as O(1) operations.
Additionally, a request with a target service time higher than the preconfigured time limit can be added to a list of outliers, e.g. the outlier list 504. Since the scheduling of these requests may not be time critical, the system permits the slower addition to a sorted list of outliers 504. Furthermore, the concurrent priority queue 501 can use a sequencer, e.g. outliers_seq, to enforce a first-in-first-out (FIFO) order for the outlier list with the same QoS.
For example, the calendar ring 502 can be configured to store requests with a target response time (or QoS) below 2 seconds, since the requests with the QoS higher than 2 seconds can be considered rare. Furthermore, the requests with the QoS below 2 seconds can be placed in the calendar ring 502 that matches QoS, while the requests with the QoS higher than 2 seconds can be placed into the list of outliers 504.
Unlike the calendar queue as shown in
Using continuation-passing, the system can transform the calendar queue 501 from non-blocking to blocking. The continuation-passing 507 can enable the consumers A-C 511-513 to manage the idle workers, or Threads 530, in the thread pool 520, so that the threads 530, which may be waiting in the thread pool 520, can be reused.
Additionally, the concurrent priority queue 501 can include a sequencer 503 that enables the concurrent priority queue 501 to detect contention and can use a fast lane 505 to support cooperative concurrency. Thus, the concurrent priority queue 501 can be aware of and handle the contention properly, without a need for the locks to expose knowledge about contention.
Add a Request into the Concurrent Priority Queue
A caller, e.g. deserializer A 611, can add an incoming request, e.g. request A 631, into the priority queue 601. For example, when the incoming request has a QoS within a preconfigured time limit, the deserializer A 611 can add the request A 631 into the calendar ring 602 directly. Otherwise, the caller can add the incoming request B 632, into the outlier list 603, when the QoS associated with the request is beyond the preconfigured time limit.
Additionally, when the priority queue 601 is empty, the priority queue 601 can wake up a thread, e.g. thread A 621, in the thread pool 610 (also referenced to as a waiter list or waiter pool). Then, a consumer, e.g. consumer A 612 that is associated with the thread A 621, can be aware of the addition of the request A 631 in the calendar ring 602 and proceed to consume the request A 631 in a timely fashion.
In accordance with one embodiment of the invention, the concurrent priority queue can use the continuation passing feature to allow a caller to manage thread in the thread pool 610. Additional information about continuation passing is disclosed in pending U.S. Patent Application titled “SYSTEM AND METHOD FOR TRANSFORMING A QUEUE FROM NON-BLOCKING TO BLOCKING”, application Ser. No. 13/781,500, filed Feb. 28, 2013.
Claim a Request from the Concurrent Priority Queue
A bit map 705 can be created and updated to be in synchronization with the calendar ring 702. When consumer A 711 tries to claim request A 731 from the calendar ring 702, the bit map 705 can be used for scanning the calendar ring 702 in order to improve efficiency of the calendar ring 702. For example, instead of examining references in the calendar ring 702, a bit scan operation on a simple bit map can be used to analyze a large number of entries (e.g. 64 entries) in the calendar ring 702.
Furthermore, the concurrent priority queue 701 can promote one or more requests, e.g. request B 732, from the outlier list 703 into the calendar ring 702, when the calendar ring 702 is empty.
Additionally, when the workload is intensive, there is a good chance that multiple requests can fall into the same calendar entry, or bucket, in the calendar ring 702. A list of stolen requests 706 can be used to reduce contention caused by claiming such requests in the calendar ring 702. Furthermore, the list of stolen requests 706 can become longer when more requests land on the same calendar entry. Thus, by taking into the list of stolen requests 706 a list of requests fallen into a single entry in the calendar ring 702 as a whole, the consumers can serialize access to the calendar ring 702 less frequently.
In accordance with various embodiments of the invention, the concurrent priority queue 702 can maintain the fast lane 704, which can be used for supporting cooperative concurrency.
Cooperative Concurrency
A sequencer 804 can be used by a victim, e.g. consumer A 811, to detect a contender, e.g. consumer B 812, for the priority queue 801. Additional information about using a sequencer in a priority queue is disclosed in U.S. Patent Application titled
“SYSTEM AND METHOD FOR USING A SEQUENCER IN A CONCURRENT PRIORITY QUEUE”, application Ser. No. 13/781,493, filed Feb. 28, 2013, and subsequently granted as U.S. Pat. No. 9,110,715 issued on Aug. 18, 2015.
In accordance with various embodiments of the invention, the priority queue 801 can reduce contention between the consumers A-B 811-812, by employing a cooperative concurrency strategy.
As shown in
Furthermore, when there are multiple contenders, the victim 811 can control the accessibility of requests in the fast lane 803 by controlling the maximum number of contenders that may be released. This is because a request can be claimed by a contender from the calendar ring 802 at a position, fastLane_w, only when the contender can set an index, fastLane_r, equal to fastLane_w (fastLane_r=fastLane_w). Since each contender may only increase fastLane_r by one, the contenders can increase the value of fastLane_r only by the number of contenders that are released.
Using the cooperative concurrency, the concurrent priority queue 801 can improve wait times, e.g. reducing the average wait time for various contenders. Additionally, the cooperative concurrency can improve CPU cache locality of the concurrent priority queue 801, since one victim consumer can do more work single-threadedly.
At step 902, if the fast lane is not empty, the consumer can attempt to claim a request from the fast lane. If the claim of a request from the fast lane fails, then there is another consumer that has successfully claimed the request. Eventually the consumer can either claim a request, or observes the fast lane is empty, in which case there are no more consumers in delete_min.
At step 903, if the fast lane is empty, the caller can proceed to the loop for claiming a request from the calendar queue. The loop for claiming a request from the calendar queue can include multiple iterations. At the beginning of each iteration, the caller can try to locate the first list of requests in the calendar array, and the caller can attempt to take the first request from the list.
At step 910, the caller can check whether there is a contention, e.g. by checking whether the readers ticket in the sequencer has been advanced beyond the last known free value at the sequencer.
Then, at step 913, if the caller detects no contention, then the consumer can proceed to claim the request from the calendar queue.
Otherwise, at step 911, if the caller detects contention, the consumer can cooperatively place the request in the fast lane, and can release one contender. As long as this consumer is inside delete_min, the released contender can eventually claim a request from the fast lane.
Furthermore, at step 912, since the request is given away to a contender, the consumer attempts to claim the next request in the calendar ring. This may lead the consumer to the step 910, which is at the beginning of another iteration in the loop.
Eventually, either the queue becomes empty, or no contenders are detected, or the fast lane is deemed full. When the fast lane is deemed full, the consumer can break the loop, update the size of the queue, and release one more contender, which will be able to eventually enter the loop for claiming requests outside the fast lane.
The present invention may be conveniently implemented using one or more conventional general purpose or specialized digital computer, computing device, machine, or microprocessor, including one or more processors, memory and/or computer readable storage media programmed according to the teachings of the present disclosure. Appropriate software coding can readily be prepared by skilled programmers based on the teachings of the present disclosure, as will be apparent to those skilled in the software art.
In some embodiments, the present invention includes a computer program product which is a storage medium or computer readable medium (media) having instructions stored thereon/in which can be used to program a computer to perform any of the processes of the present invention. The storage medium can include, but is not limited to, any type of disk including floppy disks, optical discs, DVD, CD-ROMs, microdrive, and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic or optical cards, nanosystems (including molecular memory ICs), or any type of media or device suitable for storing instructions and/or data
The foregoing description of the present invention has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations will be apparent to the practitioner skilled in the art. The embodiments were chosen and described in order to best explain the principles of the invention and its practical application, thereby enabling others skilled in the art to understand the invention for various embodiments and with various modifications that are suited to the particular use contemplated. It is intended that the scope of the invention be defined by the following claims and their equivalence.
Number | Name | Date | Kind |
---|---|---|---|
5109384 | Tseung | Apr 1992 | A |
6449614 | Marcotte | Sep 2002 | B1 |
6874144 | Kush | Mar 2005 | B1 |
6895590 | Yadav | May 2005 | B2 |
6938085 | Belkin et al. | Aug 2005 | B1 |
7046676 | Goetzinger et al. | May 2006 | B2 |
7554993 | Modi et al. | Jun 2009 | B2 |
7685391 | Cholleti et al. | Mar 2010 | B1 |
7761617 | Seigneret et al. | Jul 2010 | B2 |
7876677 | Cheshire | Jan 2011 | B2 |
7991904 | Melnyk et al. | Aug 2011 | B2 |
8130776 | Sundararajan | Mar 2012 | B1 |
8131860 | Wong et al. | Mar 2012 | B1 |
8255914 | Joyce et al. | Aug 2012 | B1 |
8347302 | Vincent et al. | Jan 2013 | B1 |
8504691 | Tobler et al. | Aug 2013 | B1 |
8539486 | Cain et al. | Sep 2013 | B2 |
8578033 | Mallart | Nov 2013 | B2 |
8850441 | Allen | Sep 2014 | B2 |
8863136 | Allen | Oct 2014 | B2 |
8918791 | Chudgar et al. | Dec 2014 | B1 |
8930584 | Otenko et al. | Jan 2015 | B2 |
20010034753 | Hildebrand et al. | Oct 2001 | A1 |
20020143847 | Smith | Oct 2002 | A1 |
20020174136 | Cameron et al. | Nov 2002 | A1 |
20030014480 | Pullara et al. | Jan 2003 | A1 |
20030053469 | Wentink | Mar 2003 | A1 |
20030078958 | Pace et al. | Apr 2003 | A1 |
20030081544 | Goetzinger et al. | May 2003 | A1 |
20030110232 | Chen | Jun 2003 | A1 |
20030120822 | Langrind et al. | Jun 2003 | A1 |
20040177126 | Maine | Sep 2004 | A1 |
20040205771 | Sudarshan et al. | Oct 2004 | A1 |
20050021354 | Brendle et al. | Jan 2005 | A1 |
20050038801 | Colrain et al. | Feb 2005 | A1 |
20050094577 | Ashwood-Smith | May 2005 | A1 |
20050102412 | Hirsimaki | May 2005 | A1 |
20050262215 | Kirov et al. | Nov 2005 | A1 |
20050283577 | Sivaram et al. | Dec 2005 | A1 |
20060015600 | Piper | Jan 2006 | A1 |
20060015700 | Burka | Jan 2006 | A1 |
20060031846 | Jacobs et al. | Feb 2006 | A1 |
20060143525 | Kilian | Jun 2006 | A1 |
20060176884 | Fair | Aug 2006 | A1 |
20060209899 | Cucchi et al. | Sep 2006 | A1 |
20060230411 | Richter et al. | Oct 2006 | A1 |
20060294417 | Awasthi et al. | Dec 2006 | A1 |
20070118601 | Pacheco | May 2007 | A1 |
20070156869 | Galchev et al. | Jul 2007 | A1 |
20070198684 | Mizushima | Aug 2007 | A1 |
20070203944 | Batra et al. | Aug 2007 | A1 |
20070263650 | Subramania et al. | Nov 2007 | A1 |
20080044141 | Willis et al. | Feb 2008 | A1 |
20080098458 | Smith | Apr 2008 | A2 |
20080140844 | Halpern | Jun 2008 | A1 |
20080286741 | Call | Nov 2008 | A1 |
20090034537 | Colrain et al. | Feb 2009 | A1 |
20090150647 | Mejdrich et al. | Jun 2009 | A1 |
20090172636 | Griffith | Jul 2009 | A1 |
20090182642 | Sundaresan | Jul 2009 | A1 |
20090327471 | Astete et al. | Dec 2009 | A1 |
20100082855 | Accapadi et al. | Apr 2010 | A1 |
20100100889 | Labrie et al. | Apr 2010 | A1 |
20100198920 | Wong et al. | Aug 2010 | A1 |
20100199259 | Quinn | Aug 2010 | A1 |
20100278190 | Yip et al. | Nov 2010 | A1 |
20110029812 | Lu et al. | Feb 2011 | A1 |
20110055510 | Fritz et al. | Mar 2011 | A1 |
20110071981 | Ghosh et al. | Mar 2011 | A1 |
20110119673 | Bloch et al. | May 2011 | A1 |
20110153992 | Srinivas et al. | Jun 2011 | A1 |
20110161457 | Sentinelli | Jun 2011 | A1 |
20110231702 | Allen et al. | Sep 2011 | A1 |
20120023557 | Bevan | Jan 2012 | A1 |
20120054472 | Altman et al. | Mar 2012 | A1 |
20120066400 | Reynolds | Mar 2012 | A1 |
20120066460 | Bihani | Mar 2012 | A1 |
20120158684 | Lowenstein et al. | Jun 2012 | A1 |
20120218891 | Sundararajan | Aug 2012 | A1 |
20120239730 | Revanuru | Sep 2012 | A1 |
20130004002 | Duchscher | Jan 2013 | A1 |
20130132970 | Miyoshi | May 2013 | A1 |
20130145373 | Noro | Jun 2013 | A1 |
20130304848 | Lyle et al. | Nov 2013 | A1 |
Number | Date | Country |
---|---|---|
2012084835 | Jun 2012 | WO |
Entry |
---|
European Patent Office, International Searching Authority, International Search Report and Written Opinion dated Mar. 14, 2014 for International Application No. PCT/US2013/067108, 12 pages. |
Baldwin, Richard G., “The ByteBuffer Class in Java”, Aug. 20, 2012, 14 pages. Retrieved from : <http://www.developer.com/author/Richard-G.-Baldwin-64720.htm>. |
European Patent Office, International Searching Authority, International Search Report and Written Opinion dated Mar. 6, 2014 for International Application No. PCT/US2013/067106, 11 pages. |
Office Action issued by United States Patent and Trademark Office for U.S. Appl. No. 13/781,500, mailed May 9, 2016 (14 pages). |
Office Action issued by United States Patent and Trademark Office for U.S. Appl. No. 14/167,792, mailed May 12, 2016 (9 pages). |
Number | Date | Country | |
---|---|---|---|
20140245312 A1 | Aug 2014 | US |