METHOD AND SYSTEM FOR SELECTION OF CLOUD-COMPUTING SERVICES
An engine for allocation of cloud computing resources to clients of cloud computing facilities (simply referenced as “clouds”) collects data relevant to resource availability and characteristics of several clouds and determines for each sought service a cloud of highest service-specific appraisal. Data corresponding to each characteristic is canonicalized to produce corresponding merits for each cloud where each merit is dimensionless, bounded within a predefined interval, and oriented so that a merit increment increases an overall appraisal of a respective cloud. Consequently, a merit vector is created for each cloud. Upon receiving a service request from a client, requisite resources as well as a significance vector indicating significance of each characteristic to the requested service are determined. An appraisal of a cloud is determined as a dot product of the significance vector and the merit vector of the cloud. The cloud of highest appraisal is allocated to the service.
The present application claims the benefit of provisional application 62/722,587, entitled “Method and System for Intelligent placement of services within a Hybrid Cloud Environment”, filed Aug. 24, 2018, the entire content of which is incorporated herein by reference.
FIELD OF THE INVENTIONThe present invention is directed towards optimal selection of clouds for clients seeking cloud-computing services.
BACKGROUNDA hybrid cloud has a number of private clouds (based on virtualization techniques) or public clouds that host services required by an organization. Public clouds offer cost-effective computing, storage, and other services such as analytics or content delivery, while private clouds deliver unrivaled control and security to meet business demands. Multiple public cloud providers are often used in a single hybrid cloud, with varying prices and ever-expanding capabilities. Private data centers frequently incorporate many different types of hardware providing a wide variety of capabilities. Different types of public and private clouds require different placement rules. There is no one-to-one mapping between the capabilities of different environments, making their comparison difficult. Deployment environments are often shared among many tenants, leading to an environment that is in constant flux in terms of capability and capacity. Placement of a service must balance out an arbitrary number of factors, such as cost, compatibility, capacity. The needs of different services vary widely, and the types of placement requirements also vary widely from one organization to another.
There is a need, therefore, to explore means for automatically matching clouds to sought services taking into account varying properties of available clouds.
SUMMARYIn accordance with an aspect, the invention provides a method of allocating cloud computing resources to clients of cloud computing facilities. The method is implemented using a placement engine having at least one hardware processor coupled to memory devices storing processor executable instructions causing the at least one processor to perform processes of collecting data relevant to resource availability and characteristics of several clouds and determining for each sought service a cloud of highest service-specific appraisal.
Initially, the engine performs a process of acquiring from each cloud of a plurality of clouds data defining available resources as well as individual valuations of characteristics of a set of predefined characteristics. For each characteristic of the set of predefined characteristics, the engine performs a process of canonicalization of valuations of the each characteristic corresponding to all clouds to determine corresponding merits where each merit is dimensionless, bounded within a predefined interval, and oriented so that an increment of the each merit increases an appraisal of a respective cloud.
Upon receiving, from a client, a service request specifying a service type of a set of predefined service types, the engine performs a process of acquiring data defining requisite resources for the service type as well as a significance vector comprising an indicator of significance of each characteristic to the service type.
An appraisal of each cloud having available resources to provide the requisite resources is then determined as a function of respective merits and the significance vector. The engine instructs the client to direct the service request to a cloud of highest appraisal.
According to a preferred implementation of the aforementioned function, all merits determined according to the canonicalizing process are organized into a plurality of merit vectors, each merit vector comprising merits corresponding to: a respective cloud; and each of the predefined characteristics considered in a predetermined order. The function is then implemented as a dot product of a merit vector of a cloud and the significance vector.
According to a first implementation of the canonicalization process, the characteristics are considered one at a time, and for each characteristic, corresponding valuations for all of the clouds are examined to determine a respective minimum valuation Xmin and a respective maximum valuation Xmax for each characteristic under consideration.
The set of predefined characteristics may comprise at least one type-1 characteristic where increasing a respective valuation increases a respective cloud appraisal. For a specific type-1 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=(x−Xmin)/(Xmax−Xmin).
The set of predefined characteristics may comprise at least one type-2 characteristic where decreasing a respective valuation increases a respective cloud appraisal. For a specific type-2 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=(Xmax−x)/(Xmax−Xmin).
According to a second implementation of the canonicalization process, the characteristics are considered one at a time, and for each characteristic a respective predefined reference valuation Xref is acquired.
For a specific type-1 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=x/(x+Xref).
For a specific type-2 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=Xref/(x+Xref).
According to a third implementation of the canonicalization process, the characteristics are considered one at a time, and for each characteristic, a cumulative distribution of corresponding valuations for all of the clouds is generated.
A respective valuation lower bound Vmin corresponding to a predefined value a1 of the cumulative distribution is determined, and a respective valuation upper bound Vmax corresponding to a predefined value a2 of the cumulative distribution is determined, 0.0<a1<a2<1.0.
For a specific type-1 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=0.0 for x<Vmin;
μx=(x−Vmin)/(Vmax−Vmin), for Vmin≤x≤Vmax
μx=1.0for x>Vmax.
For a specific type-2 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=1.0 for x<Vmin;
μx=(Vmax−x)/(Vmax−Vmin), for Vmin≤x≤Vmax
μx=0.0for x>Vmax.
Embodiments of the present invention will be further described with reference to the accompanying exemplary drawings, in which:
Cloud: A computing facility that provides computing resources on demand is conventionally referenced as a cloud.
Characteristic: The characteristic of a cloud is an attribute of the cloud, such as capacity, or am implication of engaging a cloud, such as cost.
Valuation: A valuation is a magnitude of a characteristic, in other words, valuation is a (numerical) measure of a characteristic.
Type-1 characteristic: A characteristic an increment of valuation of which increases an overall appraisal of a respective cloud is a type-1 characteristic. Processing capacity is a type-1 characteristic.
Type-2 characteristic: A characteristic a decrement of valuation of which increases an overall appraisal of a respective cloud is a type-2 characteristic. Delay is a type-2 characteristic and service cost is a type-2 characteristic.
Canonicalization: Canonicalization is a transformation that combines normalization and rectification of valuation to produce a corresponding canonical merit that is:
-
- (1) bounded within a predefined interval, preferably the closed interval [0.0, 1.0];
- (2) dimensionless; and
- (3) uniformly oriented (rectified) so that an increment of a merit increases an overall appraisal of a respective cloud whether the merit represents a type-1 characteristic or a type-2 characteristic.
Merit: A canonical merit resulting from canonicalization of valuation od a characteristic is also referenced as “merit” for brevity.
Dot product: The dot product used in the present specification is consistent with the formal definition where the dot product of a first vector {x0, x1 . . . . , x(n-1)} and a second vector {y0, y1, . . . , y(n-1))} is a scalar [x0×y0+x1×y1+ . . . +x(n-1)×y(n-1)], n>1.
Processor: The term refers to a hardware device (a physical processing device) which typically accesses at least one memory device storing processor executable instructions.
REFERENCE NUMERALS
- 100: System for educated automatic selection of cloud computing servers comprising a plurality of placement engines
- 102: A plurality of clients of a plurality of computing facilities (a plurality of “clouds”)
- 110: An individual client
- 112: A plurality of clouds
- 120: An individual cloud
- 122: Metadata and bulk data exchanged between the plurality of clients and the plurality of clouds
- 150: A plurality of placement engines
- 160: An individual placement engine
- 162: Service definition communicated to a placement engine of a plurality of placement engines
- 164: Cloud discovery data exchanged between
- 168: Placement recommendation sent from a placement engine to a client
- 220: Cloud characterization module
- 240: Storage medium holding cloud-characterization data
- 260: Cloud selection) recommendation module
- 300: Arrangement where multiple cloud-recommendation modules contend for access to a single storage medium 240
- 600: Distributed system for enabling a plurality of clients to automatically select respective clouds
- 620: A global network interconnecting clients, clouds 120, storage media 240, placement engines 160, cloud-characterization modules 220, and cloud recommendation modules 260.
- 700: Overview of the functions of a placement engine 160
- 720: Service definition module
- 730: Resource availability data
- 740: Cloud-valuation data
- 750: Request for cloud service received from a client 110
- 770: Data identifying required resources
- 780: Data identifying relevance of requested service to cloud characteristics
- 800: Example of a cloud selection module 260 communicating with a set of clouds within the distributed system 600 for cloud characteristics acquisition
- 850: Communication paths through the network connecting a cloud characterization module to a plurality of clouds
- 900: Example of placement engine 160 communicating with a set of clouds within the distributed system 600 for cloud characteristics acquisition
- 950: Communication paths through the network connecting a placement engine to a plurality of clouds
- 1000: Example of interaction of clients 110, a cloud selection module, and a storage medium
- 240 to determine preferred clouds
- 1010: Path through a network
- 1100: Example of interaction of clients 110, a placement engine, and a storage medium 240 to determine preferred clouds
- 1110: Path through a network
- 1210: Resource type
- 1220: Nominal resource-allocation threshold
- 1240: Current (time-varying) resource-allocation threshold
- 1310: Cloud characteristic index
- 1320: Valuation of a specific characteristic of a specific cloud
- 1400: Merit vectors based on cloud-valuation data of
FIG. 13 - 1440: Merit vector of a specific cloud
- 1710: Valuation matrix
- 1720: merit matrix
- 1740: Exemplary valuations of cloud characteristics
- 1750: Merit vectors derived from valuations 1740
- 1820: List of nominal resource requirements for a specific service type
- 1840: List of current resource requirements for a specific service type
- 1900: Data used for determining eligible clouds for a specific service
- 2000: Significance vectors corresponding to service types for a specific client or a specific client group
- 2020: Significance coefficient of a cloud coefficient with respect to a specific service type for a specific client
- 2040: A vector of significance coefficients with respect to a specific service type for a specific client or a specific client group
- 2120: Significance coefficient of a cloud coefficient with respect to a specific client, independent of service type
- 2140: A vector of significance coefficients with respect to a specific client
- 2200: Matrix of service-cloud compatibility
- 2210: A compatible cloud for a specified service
- 2220: An incompatible cloud for a specified service
- 2300: Client-cloud distance matrix
- 2310: Client index
- 2320: Distance from a client device to a specific cloud
- 2400: Processes of determining cloud merit vectors
- 2500: Processes of cloud selection
A conventional cloud-computing system enables a community of clients to communicate with clouds (computing facilities) to request services requiring web services, data storage, and various levels of data processing. A client may send a request for service to a selected cloud specifying service requirements. The client and the selected cloud exchange data to establish a service session.
Typically, individual clouds have different processing capabilities, storage capacities, and networking features. A client may be an individual user or a business organization. A business organization may request services of different types with varying degrees of resource requirements and service-quality requirements. A client may engage any cloud of a respective designated subset of clouds and initiate service sessions as the need arises. Consequently, any cloud may be actively providing service to several clients concurrently. Naturally, the clients' activities are uncoordinated. Thus, while the combined provisioned resources of client-accessible clouds may exceed the overall resource requirements of the entire community of users, the fluctuating resource occupancies of the individual clouds may lead to several clouds being fully occupied while, concurrently, other clouds have significant resource vacancies. A client may use a cloud-monitoring tool to find a cloud having sufficient free resources for a specific service.
The present invention introduces a placement engine configured to receive service requests from clients and determine, for each service request, an available cloud of highest merit measure.
The term “client” is used herein to refer to a communication device configured to communicate with individual clouds 120 and with individual placement engines 160. Generally, a business organization may employ multiple communication devices (multiple clients) to interact with the clouds 120. The plurality 150 of placement engines 160 receive service requests from a plurality 102 of clients 110. Each service request details a service definition 162 indicating resource requirements and service-quality expectation.
Each cloud 120 is a hardware entity. The clouds 120 are naturally geographically distributed. In fact, a single cloud 120 may employ a geographically distributed data center. Likewise, the clients 110 are generally geographically distributed. The task of characterizing the clouds 120 may be divided among the placement engines 160 so that each placement engine communicates with a respective subset of the clouds 120. The characterization information may then be pooled.
A placement engine 160 captures clouds' information and service requirements for educated automatic selection of cloud computing servers. A client sends a service request to a selected placement engine 1560. Upon receiving a placement recommendation 168 identifying a preferred cloud for the requested server, the client exchanges data 122 with the preferred cloud; the data may include metadata as well as bulk data.
Unit 420 is configured to acquire a list of encoded definitions of resource types and a list of encoded definitions of cloud characteristics of interest.
Unit 440 is configured to monitor the plurality 112 of clouds 120 to acquire information relevant to provisioned resources and time-varying resource availability of each cloud 120, as well as data characterizing the clouds individually.
Unit 460 is configured to convert the acquired cloud characterization data into a canonical form where a characteristic of a cloud is expressed as a dimensionless “merit” having a value bounded between predefined limits. The predefined limits are preferable 0.0 and 1.0. The merits corresponding to different characteristics may be defined to: consistently trend towards a sought optimum value as the magnitude of a merit increases; or consistently trend towards a sought optimum value as the magnitude of a merit decreases. Without loss of generality, the former is used throughout the description below. For example, the cost of service and speed of processors may be selected as two of the characteristics of clouds. A high cost is a disadvantage while a high processing speed is an advantage. Using cost as the sole criterion for selecting a preferred cloud 120 of the plurality of clouds, the cloud corresponding to minimum cost would be selected. Using processing speed as the sole criterion, the cloud corresponding to highest processing speed would be selected. Since both low cost and high processing speeds are desirable characteristics, but a cloud providing the lowest cost may not provide the highest processing speed of all clouds that are available for a specific service, the clouds 120 of the plurality 112 of clouds may be individually appraised according to a weighted sum of respective cloud valuations with respect to service cost and processing speed.
With an arbitrary number of predefined characteristics, with some characteristics, such as cost and delay, trend towards optimality as their respective values decrease, while other characteristics, such as processing speed, memory speed, and memory capacity, trend towards optimality as their respective values increase, a canonical representation of the characteristics need be explored. The sought canonical representation would also circumvent the difficulty of comparing clouds arising from characteristics' measurement units of differing dimensions and widely differing quantifications.
As mentioned above, service cost may be selected as one of the clouds characteristics. A service, however, may comprise multiple service facets with the service cost itemized for allocation of each facet. To facilitate cloud valuation based on cost, according to an embodiment of the present invention, a “service basket” or a “service bundle” is used a service unit for costing purposes. The service unit is a predefined list of service items of predefined proportions to be used for evaluating the cost of service of a specific cloud.
Unit 460 structures the merits of each cloud with respect to each characteristic to form merit vectors. Each merit vector corresponds to a respective cloud and comprises a merit value (a scalar) for each characteristic according to a predefined order.
Unit 480 is configured to organize storage medium 240 to facilitate insertion and retrieval of the resource availability data acquired in unit 440 and merit vectors generated in unit 460.
Unit 520 is configured to acquire for each registered client (i.e., a client that has previously engaged the cloud-recommendation module) a respective significance vector indicating significance of each cloud characteristic, based on the list of encoded definitions of cloud characteristics of interest, to each service type of a predefined list of service types. Unit 520 assembles the information in a suitable data structure to facilitate insertion, update, and retrieval of the significance indicators. The table of
Unit 540 is configured to receive a service request from a client where the request indicates a service type and requisite resources of a predefined list of resource types. If the client is a registered client, information specific to the client of significance of individual cloud characteristics to the service type may be retrieved from memory device 280. Otherwise, the received service request may explicitly define relevance of the service to each cloud characteristic based on the list of encoded definitions of cloud characteristics of interest. Such information is then added to the characteristics' significance data (memory 280).
Unit 550 is configured to compare the requisite resources with the available resources, which generally vary with time) of each cloud 120 of the plurality 112 of clouds. A set of eligible clouds, each of which having sufficient available resources, is considered for placing the sought service.
Unit 560 is configured to determine a dot product of a significance vector of a specific service and a merit vector of a specific cloud. The value of the dot product (a scalar) of the two vectors is a measure of service advantage of engaging the specific cloud to provide the sought service. Thus, for the received service, unit 560 determines a significance vector. Unit 560 then accesses storage medium 240 to retrieve a merit vector for each cloud of the set of eligible clouds, and computes a dot product of the significance vector and each retrieved merit vector.
Unit 580 communicates with the requesting client to recommend installing the service at the cloud corresponding to the highest dot product.
The cloud-characterization module 220 performs the functions of:
-
- (1) acquiring cloud information relevant to resource availability and cloud-characteristic valuation;
- (2) updating the content of cloud-characteristics storage medium 240 as the need arises;
- (3) computing new merit vectors; and
- (4) updating existing merit vectors.
The cloud-characterization module 220 stores the resource availability data, the cloud-characteristic valuation data, and the merit vectors in cloud-characteristics storage medium 240.
The cloud-selection module 260 acquires resource-availability data 730 as well as cloud valuation data 740 from storage medium 240.
The network interface 210 comprises a service-definition module 720 which receives clients' service requests 750 and formulates, for each request, service-definition data which includes data 770 identifying required resources, and data 780 identifying relevance of requested service to cloud characteristics.
As illustrated in
The characteristic of index 0 represents service cost, denoted c. The valuations c for the 12 clouds (K=12) are denoted c0, c1, . . . , c11. The valuations are canonicalized to produce corresponding merits denoted α0,0, α1,0, . . . , α11,0, as indicated in
The characteristic of index 2 represents available processing capacity, denoted p. The valuations of p for the 12 clouds (K=12) are denoted p0, p1, . . . , p11. The valuations are canonicalized to produce corresponding merits denoted α0,2, α1,2, . . . , α11,2.
The characteristic of index 5 represents service delay, denoted d. The valuations of the service delay for the 12 clouds (K=12) are denoted d0, d1, . . . , d11. The valuations are canonicalized to produce corresponding merits denoted α0,5, α1,5, . . . , α11,5.
The characteristic of index 8 represents available storage capacity, denoted q. The valuations of q for the 12 clouds (K=12) are denoted q0, q0, . . . , q11. The valuations are canonicalized to produce corresponding merits denoted α0,8, α1,5, . . . , α11,8.
Characteristics c and d (columns 0 and 5 of the matrix of
Characteristics p and q (columns 2 and 8 of the matrix of
Merit vector 1440a appraises the cloud of index 0. The elements {α0,0, α0,1, α0,2, α0,3, α0,4, α0,5, α0,6, α0,7, α0,8} are canonicalized values of corresponding valuations of row 0 (cloud of index 0) of
The elements {α11,0, α11,1, α11,2, α11,3, α11,4, α11,5, α11,6, α11,7, α11,8} are canonicalized values of corresponding valuations of row 11 (cloud of index 11) of
As described above, with reference to
A valuation, x, of a cloud characteristic may vary significantly between clouds 120. Depending on the characteristic type, the overall advantage of the cloud may increase as x increases or as x decreases. To facilitate consideration of multiple characteristics of different types, the valuation x is canonicalized, where x is converted into a merit μx which consistently trend towards a sought optimum value. Several implementations of such canonicalization may be considered.
According to one embodiment of the canonicalization process, illustrated in
According to a first implementation of the canonicalization process, the characteristics are considered one at a time, and for each characteristic, corresponding valuations for all of the clouds are examined to determine a respective minimum valuation Xmin and a respective maximum valuation Xmax for each characteristic under consideration.
The set of predefined characteristics may comprise at least one type-1 characteristic where increasing a respective valuation increases a respective cloud appraisal. For a specific type-1 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=(x−Xmin)/(Xmax−Xmin).
The set of predefined characteristics may comprise at least one type-2 characteristic where decreasing a respective valuation increases a respective cloud appraisal. For a specific type-2 characteristic, of a specific cloud, the characteristic having a valuation x, a corresponding merit is determined as:
μx=(Xmax−x)/(Xmax−Xmin).
The processing-capacity valuation, denoted p, for the plurality 112 of clouds, varies between Pmin and Pmax. A processing-capacity merit, μp, is defined as:
μp=(p−Pmin)/(Pmax−Pmin).
Thus, a cloud having a processing-capacity valuation p equal to Pmin, is given a merit of 0.0 while a cloud having a processing-capacity valuation of Pmax is given a merit of 1.
Likewise, the storage-capacity valuation, denoted q, for the plurality 112 of clouds, varies between Qmin and Qmax. A storage-capacity merit, μq, is defined as:
μq=(q−Qmin)/(Qmax−Qmin).
Thus, a cloud having a storage-capacity valuation q equal to Qmin, is given a merit of 0.0 while a cloud having a processing-capacity valuation of Qmax is given a merit of 1.
For each of the two characteristics, the valuation trends towards optimality (increasing the overall advantage of a respective cloud) as the valuation increases.
The service-cost valuation, denoted c, for the plurality 112 of clouds, varies between Cmin and Cmax. A service-cost merit, μc, is defined as:
μc=(Cmax−c)/(Cmax−Cmin).
Thus, a cloud having a service-cost valuation c equal to Cmin, is given a merit of 1.0 while a cloud having a cost valuation of Cmax is given a merit of 0.
Likewise, the estimated service-delay valuation, denoted d, for the plurality 112 of clouds, varies between Dmin and Dmax. A service-delay merit, μd, is defined as:
μd(Dmax−d)/(Dmax−Dmin).
Thus, a cloud having a service-delay valuation d equal to Dmin, is given a merit of 1.0 while a cloud having a cost valuation of Dmax is given a merit of 0.
For each of the two characteristics, the valuation trends towards optimality (increasing the overall advantage of a respective cloud) as the valuation decreases.
According to a variation of the method illustrated in
Thus, for each characteristic, a cumulative distribution of valuations over all of the clouds of the plurality of clouds is generated. A valuation lower bound Vmin of a specific characteristic corresponds to a predefined value a1 of the cumulative distribution. A valuation upper bound Vmax of the specific characteristic corresponds to a predefined value a2 of the cumulative distribution, 0.0<a1<a2<1.0.
A set of predefined characteristics may comprise at least one type-1 characteristic where increasing a respective valuation increases a respective cloud appraisal. A merit corresponding to a specific type-1 characteristic for a specific cloud having a valuation x is then determined as:
μx=0.0 for x<Vmin;
μx=(x−Vmin)/(Vmax−Vmin), for Vmin≤x≤Vmax
μx=1.0for x>Vmax.
The set of predefined characteristics may comprise at least one type-2 characteristic where decreasing a respective valuation increases a respective cloud appraisal. A merit corresponding to a specific type-2 characteristic for a specific cloud having a valuation x is determined as:
μx=1.0 for x<Vmin;
μx=(Vmax−x)/(Vmax−Vmin), for Vmin≤x≤Vmax
μx=0.0for x>Vmax.
According to another embodiment of the canonicalization process, the merit is determined based on a predefined reference valuation of a characteristic. For valuation, x, of a specific cloud characteristic of reference valuation Xref, the merit is determined according to the transformation:
μx=x/(x+Xref) for a type-1 characteristic where increasing x increases the overall advantage of a respective cloud, or
μx=Xref/(x+Xref) for a type-2 characteristic where decreasing x increases the overall advantage of the respective cloud.
In either case, μx has asymptotic values of 0.0 and 1.0:
-
- (i) for a type-1 characteristic, μx tends to 0.0 as x tends to 0.0 and tends to 1.0 as x tends to infinity; and
(ii) for a type-2 characteristic, μx tends to 1.0 as x tends to 0.0 and tends to 0.0 as x tends to infinity.
Xref=64.0 units, type-1 characteristic
Xref=80.0 units, type-2 characteristic
The valuations of the four clouds according to the characteristic of index (0), which is a type-2 characteristic, are determined to be 122.4, 40.0, 240.0, and 50.9, respectively. The minimum and maximum values are 40.0 and 240.0, respectively. Thus, the corresponding canonicalized metrics are determined as (
{(240.0-122.4)/(240.0-40.0)},
{(240.0-40.0)/(240.0-40.0)},
{(240.0-240.0)/(240.0-40.0)}, and
{(240.0-50.9)/(240.0-40.0)}.
The results, 0.588, 1.0, 0.0, and 0.946 are indicated in the column of index (0) of matrix 1720.
The valuations of the four clouds according to the characteristic of index (6), which is a type-1 characteristic, are determined to be 42.9, 30.0, 80.0, and 68.2, respectively. The minimum and maximum values are 30.0 and 80.0, respectively. Thus, the corresponding canonicalized metrics are determined as (
{(42.9-30.0)/(80.0-30.0)},
{(30.0-30.0)/(80.0-30.0)},
{(80.0-30.0)/(80.0-30.0)}, and
{(68.2-30.0)/(80.0-30.0)}.
The results, 0.258, 0.0, 1.0, and 0.764 are indicated in the column of index (6) of matrix 1720.
For a service type under consideration, the service type of index (2) for example, the nominal resource requirements are denoted U2,h, 0≤h<H, and the current resource requirements are denoted u2h, 0≤h<H, the total number H of resource types being 5 in the example of
As illustrated in
To fulfil the current requirements, a cloud of index k is qualified as a candidate cloud for the requested service if u2,h≤rk,h, for each value of h.
To fulfil requirements relevant to some contractual agreements, a cloud is qualified if U2,h≤Rk,h, for each value of h.
The merit vectors 1440 (
The entries of matrix 2300 may be determined from known longitude-latitude coordinates of client premises and cloud premises. However, it may be feasible to acquire round-trip propagation delays between a client and each cloud site.
To take the distances (or propagation delays) into account, the distance (or propagation delay) may be canonicalized to determine respective merit values and a predetermined fraction of the distance merit (propagation-delay merit) may be added to the dot product of respective merit vector 1440 (
In process 2430, the at least one hardware processor quantifies, for each cloud 120 each characteristic of the list of relevant cloud characteristics to produce a characteristic valuation for each cloud-characteristic pair. The valuations are organized into a respective data structure. For ease of illustration, the characteristics valuations are organized into a matrix structure (
In process 2440, the at least one hardware processor, canonicalizes each column of the matrix to produce dimensionless cloud-characteristics merits. In accordance with an embodiment, each merit is bounded within a predefined dimensionless interval; preferable the interval [0.0, 1.0]. As described above, with reference to
Process 2510 receives a service request from a client 110, the request specifies a service type of a list of predefined service types.
Process 2520 acquires resource requirements corresponding to the requested service.
Process 2530 acquires data indicating resource availability for each cloud 120 of the plurality 112 of clouds. The data would be retrieved from storage medium 240 which maintains time-varying resource availability data for each cloud 120 of the plurality 112 of clouds.
Process 2540 identifies a set of eligible clouds of the plurality 112 of clouds, where each eligible cloud has sufficient free resources to handle the requested service.
Process 2550 acquires a significance vector corresponding to the specified service.
Process 2560 determines a dot product of the significance vector and a merit vector of each eligible cloud. The merit vectors may be read from storage medium 240.
Process 2570 recommends the eligible cloud corresponding to the highest dot product to the client.
A merit of a cloud k with respect to a cloud characteristic j is denoted αk,j. A significance coefficient of a characteristic of index j for a service type of index s, 0≤j<J, 0≤s<S, is denoted βs,j, K being the total number of clouds 120 of the plurality of clouds 120, J being the total number of cloud characteristics under consideration, and S being the total number of service types,
A merit vector of a cloud k comprises elements:
-
- {αk,0, αk,1, αk,2, . . . αk,J-2, αk,J-1}.
A significance vector of a service s (for a specific client or a specific client group) comprises elements:
-
- {βs,0, βs,1, βs,2, . . . βs,J-2, βs,J-1}.
The dot product of a merit vector of a cloud k and a significance vector of a service s, i.e., the appraisal of candidate cloud k for a service s is determined as: - αk,0×βs,0+αk,1×βs,1+αk,2×βs,2 . . . +αk,J-2×βs,J-2+αk,J-1×βs,J-1.
- {βs,0, βs,1, βs,2, . . . βs,J-2, βs,J-1}.
A service may comprise multiple components that may be directed to a single cloud or more than one cloud. Upon receiving a request for service, the placement engine may identify all clouds that can handle the requirements of the multiple components together and select a cloud of highest merit. Alternatively, the placement engine may treat the components separately and identify for each component specific clouds that can handle respective requirements and select a cloud of highest merit. Thus, the components may be treated as separate services except in the case of temporal constraints where activation of the component has to take place concurrently or within a specified time window. In this case, after ensuring concurrent availability of resources in multiple clouds, the differing propagation delays between the client's device and the individual clouds as well as the differing queueing delays at the clouds may have to be taken into consideration.
Processor-executable instructions causing respective hardware processors to implement the processes described above may be stored in processor-readable media such as floppy disks, hard disks, optical disks, Flash ROMS, non-volatile ROM, and RAM. A variety of processors, such as microprocessors, digital signal processors, and gate arrays, may be employed.
Although specific embodiments of the invention have been described in detail, it should be understood that the described embodiments are intended to be illustrative and not restrictive. Various changes and modifications of the embodiments shown in the drawings and described in the specification may be made within the scope of the following claims without departing from the scope of the invention in its broader aspect.
Claims
1. A method of allocating cloud computing resources comprising:
- configuring an engine employing a processor to perform processes of: acquiring from each cloud of a plurality of clouds data defining: available resources; individual valuations of characteristics of a set of predefined characteristics; for each characteristic of the set of predefined characteristics canonicalizing respective valuations to determine corresponding merits of said each cloud where each merit is: dimensionless; bounded within a predefined interval; and oriented so that an increment of said each merit increases an appraisal of a respective cloud; receiving from a client a service request specifying a service type of a set of predefined service types; acquiring for said service type data defining: requisite resources; and a significance vector comprising a significance indicator of said each characteristic; determining an appraisal of each cloud having available resources to provide said requisite resources as a function of respective merits and said significance vector and instructing said client to direct said service request to a cloud of highest appraisal.
2. The method of claim 1 further comprising:
- organizing all merits determined according to said canonicalizing into a plurality of merit vectors, each merit vector comprising merits corresponding to a respective cloud and each of the predefined characteristics considered in a predetermined order; and
- selecting said function as a dot product of a merit vector of said each cloud and said significance vector.
3. The method of claim 1 further comprising determining for said each characteristic over all clouds of the plurality of clouds:
- a respective minimum valuation Xmin; and
- a respective maximum valuation Xmax.
4. The method of claim 3 wherein:
- said set of predefined characteristics comprises at least one type-1 characteristic where increasing a respective valuation increases a respective cloud appraisal; and
- a merit corresponding to a specific type-1 characteristic for a specific cloud having a valuation x is determined as: μx=(x−Xmin)/(Xmax−Xmin).
5. The method of claim 3 wherein:
- said set of predefined characteristics comprises at least one type-2 characteristic where decreasing a respective valuation increases a respective cloud appraisal; and
- a merit corresponding to a specific type-2 characteristic for a specific cloud having a valuation x is determined as: μx=(Xmax−x)/(Xmax−Xmin).
6. The method of claim 1 further comprising acquiring for said each characteristic a respective predefined reference valuation Xref.
7. The method of claim 6 wherein:
- said set of predefined characteristics comprises at least one type-1 characteristic where increasing a respective valuation increases a respective cloud appraisal; and
- a merit corresponding to a specific type-1 characteristic for a specific cloud having a valuation x is determined as: μx=x/(x+Xref).
8. The method of claim 6 wherein:
- said set of predefined characteristics comprises at least one type-2 characteristic where decreasing a respective valuation increases a respective cloud appraisal; and
- a merit corresponding to a specific type-1 characteristic for a specific cloud having a valuation x is determined as: μx=Xref/(x+Xref).
9. The method of claim 1 further comprising determining for said each characteristic:
- a respective valuation lower bound Vmin corresponding to a predefined value a1 of a cumulative distribution of valuations of said each characteristic over all clouds of the plurality of clouds; and
- a respective valuation upper bound Vmax corresponding to a predefined value a2 of the cumulative distribution, 0.0<a1<a2<1.0.
10. The method of claim 9 wherein:
- said set of predefined characteristics comprises at least one type-1 characteristic where increasing a respective valuation increases a respective cloud appraisal; and
- a merit corresponding to a specific type-1 characteristic for a specific cloud having a valuation x is determined as: μx=0.0 for x<Vmin; μx=(x−Vmin)/(Vmax−Vmin), for Vmin≤x≤Vmax μx=1.0for x>Vmax.
11. The method of claim 9 wherein:
- said set of predefined characteristics comprises at least one type-2 characteristic where decreasing a respective valuation increases a respective cloud appraisal; and
- a merit corresponding to a specific type-2 characteristic for a specific cloud having a valuation x is determined as: μx=1.0 for x<Vmin; μx=(Vmax−x)/(Vmax−Vmin), for Vmin≤x≤Vmax μx=0.0for x>Vmax.
Type: Application
Filed: Aug 26, 2019
Publication Date: Feb 27, 2020
Inventors: Brian Andrew CLOW (Ottawa), Mark Ian JAMENSKY (Ottawa)
Application Number: 16/550,835