Modeling Storage System Performance for Data

Management in Cloud Environment using Ontology

Stanisław Polak

, Darin Nikolow

, Renata Słota

and Jacek Kitowski

1,2

AGH - University of Science and Technology, Institute of Computer Science,

Mickiewicza 30, 30-059 Krak´ow, Poland

ACC Cyfronet AGH, Nawojki 11, 30-950 Krak´ow, Poland

Abstract. The progress made in the ﬁeld of Cloud computing and the continu-

ously growing users demand for services with guaranteed storage performance

parameters bring new challenges. The storage system monitoring, resource sche-

duling and performance prediction are essential for successful operation of the

given distributed environment and for fulﬁllment of the Service Level Agree-

ment. Taking into account the heterogeneity of storage resources in distributed

environments it is essential to provide a transparency of monitored storage sys-

tem performance parameters. In this paper we present a common storage system

model regarding Quality of Service requirements and dynamics of performance

parameters. We also present the process of the storage ontology development

based on this model, and we show an use-case of the proposed ontology in a

storage monitoring service.

1 Introduction

As the Cloud computing paradigm gains popularity new challenges for the service

providers arise. Users have a set of requirements, described more or less formally in the

Service Level Agreement (SLA), regarding the service they demand. A subset of these

requirements may concern Quality of Service (QoS) of the storage systems where the

users keep their data, especially if the users are interested in running data oriented ap-

plications, for example, on the Grid. The storage system monitoring, resource schedul-

ing and performance prediction are essential for successful operation of the given dis-

tributed environment and for fulﬁllment of the SLA.

Modern data centers provide the infrastructure necessary for constructing the men-

tioned distributed environment [1]. Various services like SaaS, PaaS, etc. can be present

in this environment sharing the infrastructure. Since data sets are often replicated in

such environments for high availability or performance reasons, those services could

optimize their access to the data using storage performance prediction. Taking into ac-

count the heterogeneity of storage resources in distributed environments it is essential

to provide a transparency of the monitored storage system performance parameters nec-

essary for the prediction.

Using semantics for QoS description of available services allows for better service

selection [2]. Similarly, using ontologies to describe storage resources is expected to

Polak S., Nikolow D., Slota R. and Kitowski J..

Modeling Storage System Performance for Data Management in Cloud Environment using Ontology.

DOI: 10.5220/0003352100540063

In Proceedings of the International Workshop on Semantic Interoperability (IWSI-2011), pages 54-63

ISBN: 978-989-8425-43-0

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

bring more efﬁcient storage resource usage and better storage service interoperability

in distributed environment with QoS/SLA requirements.

In this paper we present the OntoStor ontology and its development method. The

OntoStor ontology describes Mass Storage Systems (MSS) like: Hierarchical Storage

Management (HSM) systems, disk arrays and local storage based MSS (local disks

attached to a server), and their QoS aspects. Our develepment method involves: 1) De-

signing the Common Mass Storage System Model (CMSSM), 2) Creating its standards-

based version called CIM based Common Mass Storage System Model (C2SM), 3)

Developing the ontology itself.

The rest of the paper is organized as follows: ﬁrst, state of the art on modeling of

storage systems is presented. In section 3 the CMSSM and the C2SM models are pre-

sented. Next, the proposed ontologies of storage resources are described. Application

of one of them is shown in section 5. The last section summarizes the paper.

2 State of the Art

There are a few popular models of information systems like: CIM, SMI-S, and GLUE.

All of them are shortly characterized below.

2.1 Common Information Model

Common Information Model (CIM) [3] is an open standard developed by the Dis-

tributed Management Task Force (DMTF). It is a hierarchical, object oriented model

of management elements in Information Technology (IT) environment. CIM as a gen-

eral model, is not bound to a particular implementation. It consists of two parts: an

Infrastructure Speciﬁcation and a Schema.

CIM Infrastructure Speciﬁcation contains a description of object oriented meta-

model based on Uniﬁed Modeling Language(UML), i.e., Meta Schema, details for inte-

gration with other models, and a grammar of Managed Object Format (MOF) language.

The basic elements of Meta Schema are: Schemas, Classes, Properties, and Methods.

Additional elements are: Indications, Associations, and References.

CIM Schema contains a set of predeﬁned classes, their properties, methods, and

dependencies among these classes. CIM Schema consists of two separated layers: Core

Model and Common Models.

Core Model deﬁnes common classes (the basic dictionary), which are used by ele-

ments of Common Models

Common Models have a set of predeﬁned models. These models are independent of

any implementation or technology, and describe particular areas of management.

CIM is the main component of Web-Based Enterprise Management (WBEM) systems

used for a distributed management of computing environments [4].

CIM has many schemas relevant to MSS: Storage Devices, Storage Services, Stor-

age Capabilities and Settings, Storage Statistics, Physical Component. Because the

HSM systems as well as most of the performance related parameter of MSS are not

directly represented in CIM we decided to build our model - C2SM.

Existing CIM schemas can be extended, and new ones can be designed. In extending

schemas, CIM speciﬁcation recommends using only classes deﬁned in the Core model.

2.2 Storage Management Initiative - Speciﬁcation

Storage Management Initiative - Speciﬁcation (SMI-S) [5] is a CIM based standard

formulated by Storage Networking Industry Association (SNIA), which deﬁnes an in-

terface to manage heterogeneous data storage environment consisting of: data storage

devices, data storage systems, and management applications. Nowadays more than 500

products are SMI-S compliant. In SMI-S, CIM classes are grouped in proﬁles (e.g. disk

array), and each of them can have subproﬁles. To be SMI-S compliant, the product ven-

dor has to implement all the mandatory CIM classes speciﬁed in the proﬁle. Subproﬁles

describe a part of management domain, and represent optional functionality of product

(e.g. client can discover a remote management interface).

In SMI-S 1.1 which is ANSI standard, there are four groups of proﬁles: Storage,

Host, Fabric Topology, and Server. The ﬁrst one has proﬁles, which directly regards to

data storage devices but they do not represent the needed performance parameters.

2.3 Grid Laboratory Uniform Environment

Grid Laboratory Uniform Environment (GLUE) [6] is a conceptual, object-oriented,

information model of Grid environments, and its main aim is to provide interoperability

among elements of the Grid infrastructure.

The basic elements of the model are main entities, which represent the core con-

cepts of the Grid environment: resource, service, location, etc. Conceptual models of

computing and storage services are deﬁned based on the main entities. These entities

are described with UML Class Diagrams. GLUE does not allow adding new classes or

associations to the model. Two approaches are available for extending the information

associated with the existing classes: placing additional information in a special attribute,

which is present in each class or creation of key/value pairs and then linking them to

the class.

In GLUE the model of MSS is deﬁned with the following classes: StorageService,

StorageServiceCapacity, StorageAccessProtocol, StorageEndpoint, StorageShare, Stor-

ageShareCapacity,StorageManager,DataStore, ComputingService.In order to have dy-

namics represented in the model it should be extended but GLUE is less extensible than

CIM, is Grid oriented, and GLUE developers plan to use CIM for the modeling of

GLUE 2 [7].

The models mentioned above do not fully address the performance aspects of MSS

which are needed for the proper management of storage in distributed environment with

QoS constraints. This fact motivated us to develop a CIM based model - C2SM.

3 Models of Data Storage Systems

In this section we present our two models describing MSS. The ﬁrst one is the Common

Mass Storage System Model (CMSSM), and the second one, the CIM based Common

mass storage System Model (C2SM), is a modiﬁed version of CMSSM arisen as a result

of taking into account the CIM standard. Due to the use of the C2SM model as a base

for the OntoStor ontology, we describe it in more detail.

3.1 The CMSSM Model

The CMSSM model describes three kinds of MSS: the HSM systems, disk array sys-

tems, and local disk systems. This model consists of sets of parameters describing the

current state of a speciﬁc MSS, its conﬁguration (e.g. total capacity of MSS, maximal

read transfer rate), and the parameters of physical devices and media (e.g. number of

tape tapes, tape block size). CMSSM uses our earlier work on modeling HSM systems

presented in [8]. The current version of CMSSM can be found in [9].

3.2 The C2SM Model

The CIM based version of CMSSM — the C2SM model consists of a set of classes

(see Fig. 1) brieﬂy described further. The AGH StorageSystem class is the main

class and it stores common information for all kinds of MSS. The AGH DiskArray-

System, AGH LocalDiskSystem and AGH HSMSystem classes represent, the

mentioned above, kinds of MSS.

The AGH DiskArraySystem contains information about disk array systems.

These systems consist of: a disk array (the CIM ComputerSystemclass) and a server

(the CIM ComputerSystem class).

The AGH LocalDiskSystem class stores information about local disk systems.

These systems comprise a server (the CIM ComputerSystem class) having hard

disks.

The AGH HSMSystem contains information common for the HSM systems. These

systems include: a server (the CIM ComputerSystem class), media libraries (the

AGH MediaLibrary class), and disks (the CIM DiskPartition class). Media

libraries contain: slots (the AGH Slot class), changer devices (the AGH Changer-

Device class), and drives (the AGH MediaAccessDevice class). The AGH HSM-

Media, AGH HSMDriveState, AGH HSMFile, and CIM LogicalFile classes

describe the states of media, drives, and ﬁles stored in HSM respectively.

4 The OntoStor Ontology

OntoStor [10] is a research project with the purpose of developing an ontology-based

methodology concerning the organization of data access in Grid environments. Differ-

ent kinds of MSS and services for monitoring and estimation of data access are semanti-

cally described. The semantic description allows for efﬁcient use of MSS and for easier

creating and integration of new Grid-enabled, data access applications.

Based on the C2SM model described above, the OntoStor ontology has been cre-

ated (See Fig. 2). At the beginning, this model was written in Managed Object For-

mat (MOF), and then converted into Web Ontology Language (OWL) format using

the ’cim2owl’ tool [11]. Next the result ﬁle was modiﬁed by a human, e.g.: redundant

Fig.1. The C2SM model.

components (individuals, properties, classes) were removed, closure axioms for some

classes were added, new properties, and individuals representing existing MSS systems

were created. As a result we received an ontology in which the classes of the C2SM

model were represented by OWL classes, and the class properties by datatype proper-

ties in OWL. For example, the ’BlockSize’ property of the AGH HSMMedia class is

represented in OWL as follows:

<owl:DatatypeProperty rdf:ID="AGH_HSMMedia__BlockSize">

<rdfs:range rdf:resource="http://www.w3.org/2001/

XMLSchema#nonNegativeInteger"/>

<rdfs:domain rdf:resource="#AGH_HSMMedia"/>

</owl:DatatypeProperty>

The three kinds of MSS mentioned in section 1 are represented by the following classes:

AGH HSMSystem, AGH DiskArraySystem and AGH LocalDiskSystem.

AGH_MediaLibrary

AGH_HSMDriveState

AGH_HSMState

CIM_PhysicalComponent

CIM_PhysicalMedia

CIM_Location

AGH_Slot

CIM_StorageLibrary

CIM_EnabledLogicalElement

CIM_LogicalElement

AGH_HSMFile

CIM_UnixFile

AGH_LocalDiskSystem

CIM_LogicalFile

CIM_ComputerSystem

AGH_StorageSystem

CIM_StorageExtent

CIM_MediaPartition

CIM_DiskPartition

CIM_ManagedElement

Thing

CIM_StatisticalInformation

AGH_DiskArraySystem

AGH_DiskBasedSystem

CIM_DeviceStatisticalInformation

CIM_ManagedSystemElement

CIM_MediaAccessDevice

AGH_ChangerDevice

CIM_PhysicalElement

CIM_Value

FunctionalObjectProperty

AGH_HSMMedia

AGH_MediaAccessDevice

CIM_ChangerDevice

CIM_MediaTransferDevice

CIM_LogicalDevice

AGH_HSMSystem

CIM_System

is−a

hasPart

contains

is−a

hasPart

is−a

hasPart

is−a

hasPart

is−a

hasPart

is−a

hasPart

is−a

hasPart

is−a

hasPart

Fig.2. The OntoStor ontology.

Their components are represented by the following classes: CIM ComputerSystem,

AGH MediaLibrary, AGH HSMMedia, and CIM DiskPartition. Components

of media libraries of the HSM system are represented by the AGH ChangerDevice

class, the AGH MediaAccessDevice class, and the AGH Slot class. The AGH-

HSMMedia, AGH HSMDriveState, AGH HSMFile, and CIM LogicalFile

classes represents information about states of media, drives, and ﬁles stored in HSM.

Using this ontology we described the resources in our testing environment, i.e., the

HP 660ex magneto-optical library and the ATL 7100 tape library, and their compo-

nents. Below we present a fragment of an OWL ﬁle containing the description of the

mentioned above, magneto-optical library.

<owl:Thing rdf:about="#Magneto_Optical_library_HP_660ex">

<rdf:type rdf:resource="#AGH_MediaLibrary"/>

<AGH_MediaLibrary__NumberOfSlots

rdf:datatype="&xsd;unsignedByte">

128

</AGH_MediaLibrary__NumberOfSlots>

<AGH_MediaLibrary__NumberOfDrives

rdf:datatype="&xsd;unsignedByte">

</AGH_MediaLibrary__NumberOfDrives>

<AGH_MediaLibrary__VendorString

rdf:datatype="&xsd;string">

</AGH_MediaLibrary__VendorString>

</owl:Thing>

As we can see this ontology, contains semantically described information about a

concrete value of a concrete parameter of MSS. Thanks to this ontology we are able to

ﬁnd a kind of MSS and their components, base on values like numbers or strings, e.g.,

”ﬁnd media libraries which have four or less drives”.

4.1 The OntoStor-ATN Ontology

The ontology described above does not cover all applications areas, e.g., it can not be

used to identify a kind of MSS based on the names of the attributes. Since this kind of

functionality was needed in a another project, an alternative version of the ontology —

the OntoStor-ATN (Attribute Name) ontology was created (see Fig. 3). This ontology

deﬁnes two main concepts:

is-a

AGH LocalDiskParameters

AGH Parameters

AGH DriveParameters

AGH DiskArrayAttributes

AGH PoolAttributes

AGH LibraryAttributes

AGH LibraryParameters

AGH DriveAttributes

AGH TapeParameters

AGH LocalDiskAttributes

Thing

AGH TapeAttributes

AGH Attributes

AGH HSMParameters

AGH HSMAttributes

AGH DiskArrayParameters

AGH ServerAttributes

AGH ServerParameters

AGH PoolParameters

Fig.3. The OntoStor-ATN ontology.

AGH Parameters — basic class for concepts representing resources. i.e. MSS and

their components

AGH Attributes — basic class for concepts representing attributes of resources

Subclasses of AGH Parameters, like e.g.: the AGH ServerParameters class,

describe concepts, which can be identiﬁed with concrete resources, e.g., server. This

ontology does not contain individuals of these concepts. Subclasses of the second main

class, i.e., AGH Attributes, deﬁne concepts of resource attributes of individual re-

sources, e.g., the AGH DiskArrayAttributesconcept represents attributes of disk

arrays. In this case, individuals of the AGH

Attributes classes are deﬁned in the

ontology, e.g., mentioned above the ’BlockSize’ property of the AGH HSMMedia class

is represent as the ’BlockSize’ individual of the OWL AGH TapeAttributesclass.

All individuals are associated with concepts of resources by the ’hasValue’ restric-

tion — below we show how the mentioned ’BlockSize’ individual is assigned to the

AGH TapeParametersclass representing tapes.

<owl:Class rdf:about="AGH_TapeParameters">

<rdfs:subClassOf rdf:resource="AGH_Parameters"/>

<rdfs:subClassOf>

<owl:Restriction>

<owl:onProperty rdf:resource="#hasAttribute"/>

<owl:hasValue rdf:resource="BlockSize"/>

</owl:Restriction>

</rdfs:subClassOf>

</owl:Class>

In the same way we assign other individuals, in result this ontology contains complete

set of concepts related with MSS and individuals representing their attributes.

5 Application of the OntoStor-ATN Ontology

The OntoStor-ATN ontology is used in the PL-Grid project [12]. The objective of this

project is to provide Polish scientists a Grid-based high performance computing envi-

ronment enabling doing e-science research.

One of the studies undertaken on PL-Grid is data management for Virtual Organiza-

tion (VO). In Grid environments, resource providing on the basis of VO is particularly

justiﬁed, because Grid applications have often high requirements in relation to hard-

ware resources. These requirements can refer to computational power and data storage

resources. For Data Oriented Applications, which are executed on Grid, the computa-

tional power is not the only requirement — providing storage resources guaranteeing

the Quality of Service (QoS) is also necessary. In heterogeneous environments, like

Grid, prediction of the performance of shared resources is a hard task. In order to meet

the QoS requirements two kinds of information are needed: information about the cur-

rent storage resource performance utilization and information about the scheduled data

transfer. To obtain this information a monitoring system taking into account the het-

erogeneity of MSS should be used. This system has to be conﬁgured automatically and

independent of the MSS being monitored and should provide a uniﬁed monitoring layer

for the QoS related parameters.

In [13] a system using the CMSSM model and the OntoStor-ATN ontology is de-

scribed. General QoS aspects of storage resources are described by different metrics.

For different kinds of resources, these metrics have different meaning, e.g., total ca-

pacity of a local disk can be received directly from their parameters, while the total

capacity of the HSM system has to be calculated as the sum of the disk cache capacity

and capacity of each medium.

The OntoStor-ATN ontology is used to describe storage resources and their at-

tributes; QoS capabilities of these resources are described by enriched version of the

QoSOnt ontology [14]. Because the resources are semantically described, it is possible

to identify the current kind of resources, based on the name of the attributes, and as a

result, the calculation of a concrete metric is also possible.

Within the PLGrid project a semantic-oriented monitoring system called the FiVO

SLAM (Service Level Agreement) monitoring system [15] has been implemented. It

is a part of the FiVO framework [16] – a system for deployment and negotiation of

dynamic VO.

6 Summary

In this paper we described the process of developing the OntoStor ontology, allowing

to describe more precisely MSS and providing performance related properties. Since

this ontology does not cover all application areas, the OntoStor-ATN ontology has been

created. This ontology was used in a semantic-oriented monitoring system. The lesson

learned is that the internal ontology structure is essential in the case of representing

QoS/SLA parameters. It is important for the monitoring applications to have the pa-

rameter names represented as classes.

The ontologies presented in the paper are based on the proposed CIM based model -

C2SM, which describes storage resources and their performance parameters. By using

our model and ontologies we achieved transparency of monitored storage system per-

formance parameters and therefore the interoperability of monitoring and estimation

services is possible.

In the future we plan to extend our model and ontologies by adding a new kind of

MSS based on disk pools.

Acknowledgements

This research is partially supported by the MNiSW grant nr N N516 405535 and AGH-

UST grant nr 11.11.120.865.

References

1. Kant, K.: Data center evolution. Computer Networks 53 (2009) 2939–2965

2. Dobson, G.: Towards uniﬁed QoS/SLA ontologies. In: Proceedings of Third International

Workshop on Semantic and Dynamic Web Processes (SDWP). (2006)

3. Common Information Model (CIM) Standards: http://www.dmtf.org/standards/cim/ (last ac-

cess 15 Nov, 2010)

4. Web-Based Enterprise Management (WBEM): http://www.dmtf.org/standards/wbem (last

access 15 Nov, 2010)

5. Storage Management Initiative Speciﬁcation (SMI-S): http:// www.snia.org/tech activities/

standards/curr standards/smi (last access 15 Nov, 2010)

6. Grid Laboratory Uniform Environment (GLUE): http://forge.gridforum.org/sf/projects/

glue-wg. (last access 15 Nov, 2010)

7. Andreozzi, S., Burke, S., Field, L., Konya, B.: Towards GLUE 2: evolution of the com-

puting element information model . Journal of Physics: Conference Series 119 (6) (2008)

http://iopscience.iop.org/1742-6596/119/6/062009.

8. Nikolow, D., Słota, R., Kitowski, J.: Grid services for HSM systems monitoring. In

Wyrzykowski, R., Dongarra, J., Wa´sniewski, J., eds.: Proceedings of 7-th International Con-

ference, PPAM 2007, Gda´nsk, Poland, September 2007. Volume 4967 of LNCS., Springer

(2008) 321–330

9. Polak, S.: The OntoStor project report no. 4.5.1-v2 (2010) http://www.icsr.agh.edu.pl/

trac/ontostor/browser/CMSSM/raport.odt (in Polish).

10. The OntoStor project: http://www.icsr.agh.edu.pl/ontostor/ (last access 15 Nov, 2010)

11. The CIM2OWL tool: http://ﬁvo.cyf-kr.edu.pl:18888/space/CIM2OWL (last access 15 Nov,

2010)

12. The PL-Grid project: http://www.plgrid.pl/en (last access 15 Nov, 2010)

13. Słota, R., Nikolow, D., Skałkowski, K., Kitowski, J.: Data management with quality of

service in PL-Grid environment. In: KU KDM 2010 : third ACC Cyfronet AGH users’

conference : Zakopane March 18–19, 2010 : proceedings, ACK Cyfronet (2010) 35–46

14. Dobson, G., Lock, R., Sommerville, I.: Qosont: a qos ontology for service-centric systems.

In: 31st EUROMICRO Conference on Software Engineering and Advanced Applications.

(2005) 80–87

15. Funika, W., Kryza, B., Słota, R., Kitowski, J., Skałkowski, K., Sendor, J., Kr´ol, D.: Monitor-

ing of SLA Parameters within VO for the SOA Paradigm. In Wyrzykowski, R., Dongarra, J.,

Karczewski, K., Wa´sniewski, J., eds.: Proceedings of Parallel Processing and Applied Math-

ematics - PPAM 2009, 8th International Conference, Wroclaw, Poland, September 2009.

Volume II of LNCS., Springer (2010) 115–124

16. Kryza, B., Dutka, L., Słota, R., Kitowski, J.: Security Focused Dynamic Virtual Organiza-

tions in the Grid based on Contracts. In Cunningham, P., Cunningham, M., eds.: Collabo-

ration and the Knowledge Economy, Issuses, Applications, Case Studies. Volume 5, II., IOS

Press (2008) 1153–1160