WO2013008036A2

WO2013008036A2 - Modelling virtualized infrastructure

Info

Publication number: WO2013008036A2
Application number: PCT/GB2012/051683
Authority: WO
Inventors: Jeremy Bullock
Original assignee: Centrix Networking Limited
Priority date: 2011-07-13
Filing date: 2012-07-13
Publication date: 2013-01-17
Also published as: GB201111975D0; WO2013008036A3; GB201212555D0; GB2492899A

Abstract

The invention provides a method of modelling a to-be virtualized infrastructure so as to predict the computing resources required by a group of users, comprises the steps of: performing an assessment of computing resources consumed by the group of users by monitoring their use of their non-virtualised infrastucture; deriving at least one of: a User Template Session, a User Template Component Session and a User Template Image, in order to determine latency of computer resources available to the group of Users; determining an acceptable value of latency to the group of users; deriving a value of User concurrency for users within the user group; and deriving a base value, for attributes based upon: application input/output usage and/or CPU usage and/or network bandwidth, said base value being indicative of concurrency values for a specified resource consumed in a virtualized computing environment. Also provided by the invention is an apparatus for carrying out the method, which apparatus includes a display and a man machine interface and operates under the control of software which is performing steps as defined in the method.

Description

Modelling Virtualized Infrastructure

Field of the Invention

This invention relates to modelling virtualized infrastructure. More particularly the invention relates to a process intended to improve modelling virtualization of computer resources. Virtualization of infrastructure is a technique which optimises resources (such as processing capabilities, IOPS and memory) so as to operate them at higher average capacity.

Background

One way in which computer architectures and networks are often visualised is by way of an optimal virtualized infrastructure. This is a technique that enables designers, architects, programmers and users to visualise complex systems and applications and so encourage load sharing.

Companies and organisations possess hundreds, sometimes thousands, of data hardware resources, such as for example servers. Data servers are expensive capital equipment to purchase, to manage and to maintain. However, very rarely are these hardware resources utilised to their full (or nearly full) capacity.

In recent years so called virtualization has enabled groups of pre-defined users to pool hardware resources and so reduce capital cost has become popular. Virtualization of infrastructure is achieved by using specific routers (switches) operating under control of software, to augment a virtual infrastructure of systems. The virtual infrastructure of systems can be accessed and shared by many users, for example by way of a wide area network (WAN) connection. The users may not be located in the same building and so share remote facilities, yet their user experience is no different from someone with a direct connection to their own dedicated computer. The major difference is a saving in cost of hardware and maintenance as well as a saving in licence fees payable for use of software.

The virtual infrastructure can include collections of applications, computer devices such as man-machine-interfaces (MMIs) and server computer devices, such as routers and wide area networks (WANs). These applications are often visualised in block or functional diagram format. Other forms of virtualized infrastructure include: available network bandwidth and maximum and actual disk storage capacity and these too may be conveniently represented graphically in a number of differing formats so as to convey useful meaning to design and maintenance engineers, as well as end users.

In order to determine an optimal, virtualized infrastructure, it is necessary to use available data from sources such as existing file allocation tables - so called fat files. Fat is a computer file system architecture that is employed on many computer systems. It is not to be confused with so-called FAT memory files, which are found on many portable devices such as memory cards, (in digital cameras) as well as diskettes.

Fat files are present within existing user based infrastructure. They are employed in order to model and predict demand that is likely to be placed on new, virtualized infrastructure and they do this by providing an indication of so-called user concurrency. User concurrency is a term defining the specific usage of applications that a population of users are accessing at any instant or during any given time period. Specific software is available to perform this and operates in virtualized architectures so as to provide average concurrency as well as individual concurrency, typically for an estate of 2000 users.

Thin client infrastructure is a term that is used to describe client resources when, for example a client's requirements changes, such as when there is a shift towards a 'cloud based' architecture. Cloud computing is a technique that enables users to work in a variety of locations, by accessing remotely stored data and using remote systems, so as to perform a task that was previously done using resources at a desk. There are many advantages in cloud computing, in particular, surety of access to data (by way of alternative data paths as systems are networked), reduced purchase and operational costs, the ability to back-up data in several locations, the capabilities of rapid changes from small to large requirements (and vice versa) and of course flexibility of access with respect to user location.

A thin client is a computer or a computer program which depends heavily on some other computer (such as a remote server) to fulfil its traditional computational roles. In contrast a fat client is one whose computational requirements are designed to take on all these roles (as well as others), usually 'in-house' and almost always by itself. The exact roles assumed by the server may vary, from providing data persistence (for example, for diskless nodes) to actual information processing on a client's behalf.

Thin clients occur as components of a broader computer infrastructure, in which many clients share their computations with the same server. As such, thin client infrastructures can be viewed as the amortization of some computing service across several user-interfaces. This is desirable in contexts where individual so-called 'fat clients' have much more functionality or power than their infrastructure either requires or uses. This can be contrasted, for example, with grid computing.

The most common type of modern thin client is a low-end computer terminal which concentrates solely on providing a graphical user interface to the end-user. The remaining functionality, in particular the operating system, is provided by the server, which may be local or remote - such as is achieved with local area networks (LANs) or wide area networks (WANs). The shift from so-called 'fat' to 'thin' client infrastructure requires that network and storage requirements of such parameters as: application usage, memory usage, central processing unit (CPU) utilisation, are no longer measured in a synchronous mode, rather they are measured in an asynchronous mode. Synchronous processes operate in separate, networked devices and either depend on a common source emitting clocking pulses, provided by a transmitting device (clock); or by synchronizing bits or bit patterns embedded in a set of data.

A switching technique for telecommunication networks uses asynchronous time- division multiplexing (ATM) which encodes data into small, fixed-sized cells. This differs from networks such as the Internet or Ethernet LANs that use variable sized packets or frames.

ATM provides data link layer services that run over an open systems interconnected model (OSI) which describes a modelling technique of dividing communications channels into different levels. ATM operates over OSI layer 1 physical links. ATM has functional similarity with both circuit switched networking and small packet switched networking. This makes it a good choice for a network that must handle both traditional high-throughput data traffic (such as file transfers) and real-time, low-latency content, such as voice and video.

ATM uses a connection-oriented model in which a virtual circuit must be established between two endpoints before the actual data exchange begins. ATM is a switching protocol used over Synchronous optical networking (SONET) and synchronous digital hierarchy (SDH). SONET and SDH are standardized multiplexing protocols that transfer multiple digital bit streams over optical fibres using lasers or light- emitting diodes (LEDs). Such networks and protocols permit transmission of very large amounts of data. However, use of SONET/SDH backbones of public switched telephone networks (PSTNs) and Integrated Services Digital Network (ISDNs) has been declining in favour of internet protocols (IP). Recently the success of so-called cloud computing has impacted across small numbers of user groups, or numbers of user groups, often referred to as an estate. Practically it is the end result that is usually modelled, rather than the impact of the sum of individual computer devices.

Cloud based systems include databases, web servers business applications which can be consolidated and can run side by side - so improving efficiency. However, they also have separate operating systems - for example Windows (Trade Mark) or Linux (Trade Mark).

When virtualizing these systems it is not unknown to achieve consolation ratios in excess of 10:1 - for example 500 servers can be reduced to 50 or 60. Therefore in addition to saving resources, users can provision the server environment - from a time perspective - to install hardware, as often many of the historically achieved functions can be achieved in software.

Problems that are encountered when virtualizing systems are optimizing the available resources and ensuring that there are sufficient resources available for users in an unexpected event or non standard conditions.

Prior Art

An examples of a facility for using images created by backup software to recreate an entire machine, as it was at the point in time in the past when the backup was taken, is described in Canadian Patent Application Number CA 2 613 419 (Syncsort Inc). The facility can be extended so as to bring up a set of machines which together serve some logical business function as in a cluster or network of associated servers. The method may also be used to further extend an entire data centre from backup images. Although providing a useful tool for assisting the virtualization process the aforementioned system does not assist in the modelling of a virtualisation proposal or the real time modelling of users in a virtualized environment.

The present invention arose in an attempt to overcome the aforementioned problem and to provide an improved method of modelling virtualized infrastructure specifically for use in the design, implementation and management of cloud based systems.

Summary of the Invention

According to a first aspect of the invention there is provided a method of modelling the to-be virtualized infrastructure so as to predict the computing resources required by a group of users, comprises the steps of: performing an assessment of computing resources consumed by the group of users by monitoring their use of their non-virtualised infrastucture; deriving at least one of: a User Template Session, a User Template Component Session and a User Template Image, in order to determine latency of computer resources available to the group of Users; determining an acceptable value of latency to the group of users; deriving a value of User concurrency for users within the user group; and deriving a base value, for attributes based upon: application input/output usage and/or CPU usage and/or network bandwidth, said base value being indicative of concurrency values for a specified resource consumed in a virtualized computing environment.

Thus the invention is able to improve the design, implementation and management of cloud based systems by determining typical usage parameters based upon obtaining a resource demand for an application as exhibited by their use in the existing non-virtualised environment. Ideally the 'Base' is calculated first to generate a subset from which subsequent calculations are made. The amount of computing resources demanded by any given computer application is largely dependent on the content being loaded by the user. Thus, an instance of a user application which is loaded without any user content (a "Vanilla Application") will often be significantly lower than a user application loaded with user content.

The Base value is obtained for attributes such as: application input/output usage, CPU usage, IOPS and network bandwidth by examining all measures of "Vanilla Application" and determining the estate mean. It is a set of "tuples" (i.e. records), where each record contains instances of the attributes used to perform subsequent calculations where users deviance from the mean of Vanilla Applications can be measured.

Optionally a filtering process is used to generate the Base value. Ideally a sequence list is derived.

Preferably calculations are performed detailing concurrency values for a predefined group of variables. The variables are selected from the group comprising: User Template Sessions, User Template Component Sessions and User Template Images.

More particularly the optimising a computer processor, so as to allocate available resources comprising the steps of: performing an assessment of resources available to a user group; determining typical usage parameters based upon one or more criteria and this is ideally achieved by determining a running sum of user concurrencies. Ideally a cache device is disposed on a connection path between a user computer executing a software application and a network. The application exchanges data with a further computer via the network.

The cache device advantageously includes a cache memory and a processor. The cache device is configured to measure, by the processor, a first latency between the user computer and another computer.

Further the cache device is configured to determine an acceptable latency range based on the latency and a requirement of the software application. The cache device is further configured to measure a second latency between the user computer and the further computer.

The cache device is further configured to store, in the cache memory, a set of data transmitted from the user computer to the further computer, if the second latency is not within the acceptable latency range.

Ideally the assessment of resources includes an initial preparatory or assessment step (namely calculating a subset of data used as the basis of the remainder of the modelling process) and extracting definitions of User Templates, which define the specification or the 'shape' of the virtualized infrastructure.

Preferably a set of calculations is performed detailing concurrency values for a predefined group of variables. A concurrency value is an indication of the extent to which multiple users are accessing the same hardware or software at the same time. Such predefined variables may also include: User Template Sessions, User Template Component Sessions and User Template Images.

The method for modelling this involves two preparatory steps namely: calculation of a value of a subset of data, used as the basis for the remainder of the modelling process, and extracting definitions of User Templates. These subsets of data help define the shape of the virtualized infrastructure.

Building on this initial value, a set of calculations are performed detailing concurrency values for such variables as: User Template Sessions, User Template Component Sessions and User Template Images.

A first step in this process is generating the Base which is described below. The Base consists of a subset of usage information that is used for all subsequent User Template calculations. Usage information concerns what Executables are placed in a process on a device, and generally comprise the following data: i) An Executable Name

ii) A Command Line used to run the Executable

iii) Start and Stop Times

iv) A User Account that is responsible for placing the Executable in process

v) Other details such as I/O, CPU, Memory usage

The cache device is inferred every time a usage report is sent from a device to a central collection server. From the contents of usage information collected from devices, the Base is generated by filtering attributes, directly or indirectly associated with collected Usage information. These attributes relate to User Accounts. User Accounts are responsible for placing applications in process on a device. For example they determine which Business Group a specific User Account belongs to, and which Applications are deemed suitable for use in a virtualised computing environment. It is understood that this list may in itself be further filtered to assess alternative scenarios. A period of time is determined, within which the processes listed within the usage information must have either started or stopped.

An example of virtualisation modelling will now be described, with reference to the Figures, in which:

Brief Description of the Figures

Figure 1 illustrates two rows containing pivoted start and stop times, retaining in each new row the reference to the User Template and CPU utilisation;

Figure 2 illustrates next step in the process, namely: amalgamation of the pivoted start/stop time data into a single list;

Figure 3 illustrates generation of concurrency of consumption of physical resources (sum as percentage of usage of CPU) by calculating the running sum over these columns;

Figure 4 illustrates determining a maximum user template session of concurrency;

Figure 5 illustrates the next stage which is extraction for each User Template/Application Component, pair start/stop times for each Application process associated with them in the Base;

Figure 6 illustrates how pivoted start/stop time data is amalgamated into a single list;

Figure 7 shows how a Running Sum is calculated which is typically a distinct combination of User Templates; Figure 8 shows how the Maximum User Template Session Concurrency is determined; and

Figure 9 is a block diagram of an apparatus for performing the method.

Detailed Description of Preferred Embodiment of the Invention

The successful implementation of a virtualized environment depends on accurate sizing and allocation of resources. Modelling existing resource demand helps to ensure that any new virtualized environment is sized according to recorded user demand, both in terms of the applications used, the content loaded into those applications and the times when those Applications and content are used.

By computing the concurrent usage of computing resources such as memory, IOPS, disk reads and writes it is possible to accurately size a to-be virtualized environment.

The outputs of a virtualization modelling exercise (either on-screen or printed) give sufficient information for the effective sizing of a virtualized environment. The key to creating these outputs is in the computation of concurrency.

An example of modelling of virtualization is described with reference to the Figures, by way of a specific example of determining concurrency, for example for a number of distributed users who are utilising a shared resource - such as 'the cloud' - to perform operations on separate documents, in a piece of application software, such as WORD (Trade Mark) using Microsoft Office (Trade Mark). Specific reference is made to Figure 9, which is a block diagram of an example of a typical distributed system that is found in many companies, organisations and offices. Steps that are performed in the method of modelling virtualized infrastructure are then described with reference to Figures 1 to 8.

Figure 9 is an overall diagrammatical view of a plurality of servers or user work stations 10, 20, 30, 40 interconnected for example by way of network cables, routers and other hardware 50. The central processor (CPU) and memory operate under control of a host operating system 70 such as LINUX and resources from the host operating system are pooled by way of virtualization software 60.

Figure 1 represents a pivot each row containing the start and stop times, in two rows, each new row the reference to the User Template.The first row in Figure 1 indicates pairs that include the start time; an indicator marking the row as a start time, and attributes recording the consumption of physical resources by a particular process. For example a particular application may demand more random access memory or processing capability than another process and these resource demands are recorded.

The second row in Figure 1 indicates the start time, an indicator marking this row as a start time, and attributes recording the consumption of physical resources by the process. It is appreciated that these attributes are now duplicated across two rows. Both sets of attributes are accorded negative values. This data is obtained and extracted for a specific end use described below.

Figure 2 illustrates the next step in the process, namely: amalgamation of the pivoted start/stop time data into a single list; and deriving a sequence list by using an event time column (which now contains both start and stop times). In the event that stop times occur before start times a special condition arises. This special condition may be flagged. The next stage of the process involves calculating a running sum. This is achieved by passing down all rows in the sequenced list, for each user template and calculating the running sum of each start/stop indicators (to calculate user concurrency). This provides an indication referred to as User Template Session Concurrency and in practical terms this indicates the number of executables concurrently in use by User Accounts associated with the User Template at the point in time of each start/stop time i.e. the User Template Session Concurrency at that instant.

It is also possible to generate concurrency of consumption of physical resources (sum as percentage of usage of CPU) by calculating the running sum over these columns in the sequenced list shown in Figure 3.

The next stage is to determine a maximum user template session of concurrency and this is shown in Figure 4. This stage is achieved for each user template and the maximum User Session Concurrency value is calculated. For each User Template, calculating maximum concurrency values for the consumption of physical resources is required so as to be able to model actual real time usage and forecast any usage trends.

User Template Component Session Concurrency is derived by generating the Usage Event List. For each User Template a list pairing is generated that references each User Template with each Application. This can be considered at whatever level of granularity and forms part of that User Template. Optionally the same Application can form part of a different User Template.

Referring now to Figure 5, the next stage is to extract, for each User Template/Application Component, pairall pairs of start/stop times for each Application process associated with them in the Base. If it is intended to calculate concurrency of consumption of physical resources, values from the Base Usage data, along with the start/stop times, are required as seen from the example in Figure 5.

The contents of the Base therefore include all attributes mentioned above that are used in a filtering process that is used to generate the Base, but include additional attributes such as application input/output usage, CPU usage and network bandwidth. Hence, the modelling process enables modelling of concurrency values for any resource consumed in a virtualized computing environment.

A number of attributes helping to describe the base have an implicit hierarchical relationship. Most obviously in an information technology or informatics (IT) environment, user accounts are often grouped into Domains. Similarly, attributes describing applications contain hierarchies.

The specific example depicted in Figure 5 shows how by pivoting each row containing the start and stop times, into two rows and by retaining in each new row the reference to the User Template and associated Components:-

The first row of these pairs includes the start time, an indicator marking this row as a start time, and attributes recording the consumption of physical resources by the process. The second row of these pair includes the start time, an indicator marking this row as a start time, and attributes recording the consumption of physical resources by the process. Optionally these attributes are now duplicated across two rows and are given negative values.

Referring now to Figure 6, which illustrates how the pivoted start/stop time data is amalgamated into a single list. Sequencing this list by the Event Time column (which now contains both start and stop times) and considering stop times before start times where they are equal, it is apparent that a value indicative of instantaneous user concurrency is obtained. Figure 7 shows how a Running Sum is calculated. This is achieved effectively by passing down all rows in the sequenced list, for each distinct combination of User Template and Component value, calculating the Running Sum of start/stop indicators (to calculate user concurrency) and providing a User Template Component Session Concurrency. It is also possible to generate concurrency of consumption of physical resources (sum as CPU) by calculating the Running Sum over these columns in the sequenced list.

Referring now to Figure 8, the Maximum User Template Session Concurrency is determined. Therefore for each User Template firstly the maximum User Session Concurrency value is calculated. Then, for each User Template, a maximum concurrency value for the consumption of physical resources can be derived and calculated. It is apparent that the method may be applied locally or over a large area network and that specific application software is capable of undertaking all or parts of the method. It is also understood that software supported on a recordable data medium is also within the scope of the invention.

Alternatively dedicated hardware may be provided for performing the method.

The invention has been described by way of example only and with reference to the Figures and it will be appreciated that variation may been made to the embodiments described It is appreciated that there are other specific techniques that may be employed and dedicated apparatus that is configured for performing the various stages of the aforementioned method, an example of which is shown in Figure 9.

Claims

1 . A method of modelling a to-be virtualized infrastructure so as to predict the computing resources required by a group of users, comprises the steps of: performing an assessment of computing resources consumed by the group of users by monitoring their use of their non-virtualised infrastucture; deriving at least one of: a User Template Session, a User Template Component Session and a User Template Image, in order to determine latency of computer resources available to the group of Users; determining an acceptable value of latency to the group of users; deriving a value of User concurrency for users within the user group; and deriving a base value, for attributes based upon: application input/output usage and/or CPU usage and/or network bandwidth, said base value being indicative of concurrency values for a specified resource consumed in a virtualized computing environment.

2. A method according to claim wherein a running sum of user concurrencies is determined.

3. A method according to claim 1 wherein a filtering process is used to generate the Base value.

4. A method according to any preceding claim wherein at least one peak concurrency value is obtained.

5. A method according to any preceding claim wherein a sequence list is derived.

6. A method according to any preceding claim wherein calculations are performed detailing concurrency values for a predefined group of variables.

7. A method according to any preceding claim wherein user concurrency is derived by generating a Usage Event List.

8. A method according to claim 7 wherein the Usage Event List includes a list pairing that is generated for each User Template with each Application.

9. A method as claimed in any preceding claim, which involves two preparatory steps namely: calculation of a value of a subset of data, used as the basis for the remainder of the modelling process, and extracting definitions of User Templates.

10. Apparatus for carrying out the method of any of claims 1 to 9 includes a display and a man machine interface and operates under control of software which is performing steps as defined in the method of any of claims 1 to 9.