A New VirtualDisk Mapping Method for theCloud Cloud Desktop StorageClient lient

Hancong Duan,X iaoqin W ang,Ping Lu,Shengm ei Luo,and Zhiyong W ang(.University of Electronic Science and Technology of China，Chengdu 60054，China;.ZTE Corporation，Nanjing 00，China)

A New VirtualDisk Mapping Method for theCloud Cloud Desktop StorageClient lient

Hancong Duan1,X iaoqin W ang1,Ping Lu2,Shengm ei Luo2,and Zhiyong W ang2
(1.University of Electronic Science and Technology of China，Chengdu 610054，China;
2.ZTE Corporation，Nanjing 210012，China)

Integration of the cloud desktop and cloud storage platform is urgent for enterprises.However，current proposals for cloud disk are not satisfactory in terms of the decoupling of virtual computing and business data storage in the cloud desktop environment.In this paper，we present a new virtual disk mappingmethod for cloud desktop storage.In Windows，compared with virtual hard disk method of popular cloud disks，the proposed implementation of client based on the virtual disk driver and the file system filter driver is available for widespread desktop environments，especially for the cloud desktop with limited storage resources.Further⁃more，our method supports customizable local cache storage，resulting in user⁃friendly experience for thin⁃clients of the cloud desktop.The evaluation results show that our virtual disk mappingmethod performs well in the read⁃write throughput of different scale files.

virtual disk;cloud desktop;file system filter driver;customizable cache storage

1 Introduction

In recent years，cloud storage has been one of themost popular internet applications with the maturity of cloud computing[1].Cloud storage[2]makes various types of network storage deviceswork together by uti⁃lizing clustering server，network transmission and distributed file system.Compared with traditional storage devices，cloud storage is low⁃cost，high security，and flexible setup of capacity for easy expansion.With the popularity ofmobile workforce and IT resource consumerization，desktop virtualization[3]as an important application of cloud computing has been devel⁃oped.

A virtualmachine(VM)which is based on thin⁃client data approach isemergingasa virtualdesktop solution for data cen⁃ters，such as Xen Desktop[4]and VMware View[5].Most desktop storage clientsof VMs run in the virtualoperating sys⁃tem，of which the computing and storage resources are tight coupling in a virtual image file stored in the supervisor operat⁃ing system.However，once the storage client in virtual desk⁃tops caches the entire user’s cloud data，the size of the virtual image file increases.This results in inflexible data storage and scalability.

In this paper，we propose a virtual disk mappingmethod for cloud desktop storage client.Thismethod isbased on the virtu⁃al disk driver and file system filter driver and provides a flexi⁃ble，user friendly cachemanagement schema in a virtual desk⁃top environment.

The remainder of this paper is organized as follows.Section 2 describes the related works;Section 3 presents the architec⁃ture of the cloud disk client in Windows，and proposes the mappingmethod based on the virtual disk drive and the file system filter driver in detail.In Section 4，we evaluate the per⁃formance of the cloud disk and compare itwith the local disk with differentdata scales.

2 Related Works

Cloud disk services have blossomed in China and else⁃where since 2012[6].These services include Dropbox，Google⁃Drive，KingSoft KuaiPan，and Baidu CloudDisk.Most of the client sides of these cloud disks are available for Microsoft Windows，IOS，and Android.Because Windows is a wide⁃spread desktop operation system with the largestnumber ofap⁃plications and the biggestmarket share，we mainly study the cloud disk client inWindows.

There are three popular technologies of the cloud disk clientinWindows:

1)web browser

For the traditional network storage service，uploading/down⁃loading files through web browsers is themain method of ac⁃cessing data.Almostall the cloud disks provideweb clients for users to access the cloud content through any operating system with web browsers.However，it is inconvenient to upload and download files frequently through web browsers among differ⁃ent user⁃terminals，such as smart phones，personal digital as⁃sistants(PDAs)and PCs.

2)synchronized directory

In the cloud disk service，data synchronization between ter⁃minals is a typical function.The cloud disk client in Windows represents a synchronized directory，such as Google Drive[7] and Baidu CloudDisk[8].Usersdrag filesbetween the synchro⁃nized directory and the local disk is used for data synchroniza⁃tion.The synchronized directory is integrated into the local sys⁃tem as the storage space of the cloud disk.However，thismeth⁃od only stores themetadata of the cloud disk user，therefore，the cloud disk is not available in an environmentwithout net⁃work.

3)virtualdisk

Virtual disk mapping supports local disk operation habits and offline operations.It is an ideal solution to cloud disk cli⁃ents.The Windows clients of Dropbox[9]and KingSoft Kuai⁃Pan[10]use virtual disk mapping.Currently，thismethod re⁃lieson the virtualhard disk(VHD)technology.VHD[11]，[12] is a virtual disk image，ofwhich the contents exactly simulate those of a hard disk by fragmenting the file contents into fixed 512 bytes records.A virtual disk based on the VHD technique replicates all the cloud disk content.After a VHD image gets mounted，it is integrated to the local system with a drive letter. Hence，the clientapplication invokes the ReadDirectoryChang⁃esW APIofWindows[13]to capture user’s operations on the virtual disk and then synchronizes data to the back⁃end cloud storage.

However，the client based on the VHD technology does notapply to the virtualdesktop environment in which the com⁃puting and storage resources ofeach user are coupled and lim⁃ited.When storing popular multimedia data，the virtual disk consumes a lotof storage space on the localhard disk，and this results in storage pressure on the localsystem with poor perfor⁃mance.Application⁃level data protection is realized by calling the specificWindows APIto captureuser’soperations.Howev⁃er，this does notsatisfy the needsof data security for enterpris⁃es.

To solve these problems，we propose a disk mapping scheme for the cloud disk client inWindows.With the custom⁃izable cache in the local system，this disk mapping scheme is available for widespread applied environments，especially for the virtual desktop storage.The data of cloud disk can be en⁃crypted at the driver level.In this way，the proposed scheme can prevent the brute⁃force attacking the cloud content cached in the localdisk.

3 The Proposed Method

3.1 System Architectureof the Cloud Disk

The proposed system architecture of the cloud disk desk⁃top clientconsistsof the cloud storage platform and variousus⁃er terminals(Fig.1).

▲Figure1.System architectureof the cloud disk client.

In Fig.1，the cloud disk desktop client typically has four layers:presentation，logic，network communication，and local cache.

The presentation layer is the user interface of the client side，andmostcloud disks present the cloud storage space asa virtual disk or a file directory on the user terminal systems for the intuitive file controlmethods.

The business rules on the logic layer are applied as data passes from the presentation layer，through the network com⁃munication and local cache layers，and back again.

In the communication layer，all the user’soperations on the virtual disk can be synchronized to the cloud storage server，and the desktop client receives update notifications from the cloud server when other user terminals change the cloud con⁃tent.

The local cache layer stores the user metadata and cloud content in the local disk.In order to adapt to different running environments，the cloud disk client has three local cache schemes:full，customizable，and no cache.

3.2 Disk Mapping Method

On the presentation layer of the cloud disk client，the cloud disk is integrated to the operation system asa local disk. In thisway，users can get localized experience of reading and writing files on the cloud disk.From a user’s perspective，there is no difference between accessing data from the cloud disk and accessing data from a localhard disk.All the user op⁃erations are automatically synchronized to the cloud server by the background application of the client.We also propose adiskmapping scheme of the cloud disk.This scheme is based on the disk drive[14]and file system filter driver[15].

3.2.1 VirtualDisk Drive

The virtual disk drivermaps the cloud storage space to the desktop system asa localvirtualdisk.The virtualdisk driv⁃er virtualizes an image file into a local disk partition which al⁃lows users to read and write.After beingmounted，the virtual disk is integrated into My Computerwith the settable drive let⁃ter，such as the H(Fig.1).TheWindows system providesma⁃ture techniques for creating and mounting the virtual disk，in⁃volving the IOCreateDisk APIs and dispatch functions in the kernelmode.User operationson the virtualdisk are sent to the file system through the I/O request packages(IRPs)controlled byWindowsNT I/OManager(Fig.2).

3.2.2 File System FilterDriver

In Windows NT drivermode[16]，every I/O request from the client process is controlled by Windows NT I/O Manager when a user operates on the cloud disk.Before sending an IRP，I/OManager checkswhether there is an additional device object driver besides the file system driver.If not，it sends the IRP to the file system driver normally;otherwise，the IRP is sent to the additional device objectdriver first.The file system filter driver is located on the file system driver or between the file system driverand storage device driver.

We design the filter driver on the file system driver(Fig. 2).The filter drivermodule intercepts the I/O requestbefore it reaches the file system.The filter driver captures I/O requests from the user’soperation on the virtualdisk and sends them to the upper network communication layer of client application. Thewhole process is shown in Fig.2.After receiving these fil⁃tered IRPs，the network communication layer optimizes them by the file operation de⁃duplication.The optimized operation setare then synchronized to the back⁃end cloud storage via the network connection.

Based on the file system filter driver，the cloud disk client can implement the transparent data encryption[17]in the ker⁃nelmode when the filter driver captures the IRPs.This effec⁃tivelyavoids the data leakageon the client.

3.3 Client-SideData-Caching Schemes

Currently，most cloud disk desktop clientsback up all the cloud disk data onto the local hard disk，so users can access the cloud data offline.The background of client automatically synchronizes data to the cloud when the application connects to the internet.Available offline operations provide a good ex⁃perience forusers.However，because some deviceshaving lim⁃ited capacity of storage and computing，the full⁃cache storage method has poor user experience due to the large storage load on the local disk.Besides full cache，no⁃cache and customiz⁃able cache schemes are introduced to fit for different running environments.Fig.3 shows the detailed processes of these three schemes.

▲Figure2.Thedisk driver and filesystem filter driver of the client.

▲Figure3.Three localcacheschemes.

The full⁃cache scheme is themain solution ofmost cloud disks at present.With this scheme，a client backups all cloud dataon the localdisk to supportofflineoperations.

With the no⁃cache scheme，a clientonly stores themeta⁃data of the cloud disk，and every data accessing ismapped to the cloud server via the network.Thin⁃clients ofmany virtual desktop are based on the no⁃cache scheme.

The size of a client cache can be set in the customizable⁃cache scheme，which is applied for different environments. The customizable⁃cache scheme takes advantage of both full cache and no cache.It supports the offline operation，and also consumes proper amount of local storage resources.Aiming at customizing the initial cache space in different running envi⁃ronments，the client based on customizable⁃cache scheme al⁃lows users to change the capacity of local cache in need when it is running.By using cache⁃replacement algorithms such as Least Recently Used(LRU)and First In，FirstOut(FIFO)，da⁃ta cached in the local disk is regularly updated to become hotspotdata.

4 Evaluation

In this section，we verify the function of the customizable⁃cache scheme.We also present a range of read⁃write bench⁃marks to show the performance of the proposed cloud disk cli⁃entand compare itwith the localhard disk.

In all the fowling test⁃bed experiments，the proposed cli⁃ent is configured in the private computer，which runs on Win⁃dows7，with dual⁃core 2.7GHz Intel Pentium processor，4GB ofmemory，and a 500GB SATA 7200 RPMhard drive.CS⁃TORE is used as the back⁃end server，which is a desktop⁃ori⁃ented distributed cloud storage platform.

4.1 CacheData Replacement

We evaluate the cache data replacementof the cloud disk client in terms of the times of cache replacement.A directory tree stored on the localhard disk is copied to the cloud disk for the simulation of user’s file operations of the cloud disk.As discussed in section 3，the previous user data cache which is least recently used is replaced until the cache capacity drops to the reserve sizeof the clientwhen the currentcachegoesbe⁃yond the allowable size.In the following test cases，we the ini⁃tial size of local cache is 100 MB.To simulate file operations of the user，we choose 200 files random ly from the file set(in⁃cludes three level of directorieswith 500 10 kB files，200 100 kB files，200 1024 kB files and 100 10MB files)stored on the local disk，copy them to the cloud disk，and estimate the cache replacement times foreach testcase.

The cache is successfully replaced when the size of cur⁃rent cache ismore than 100MB(Fig.4).In the 6 test cases of Fig.4，the average times of the cache replacement is approxi⁃mately equal to 13.The proposed customizable cachemanage⁃ment is flexible forupperapplicationsand users.

4.2 Read-Write Performance

We use IOzone[18]，a disk I/O performance test tool，in Windows.Previous studies have shown thatmost file access in⁃volves random disk accesswith small requestsizes[19].There⁃ fore，we focus on the random disk read/write and compare the throughputwith the local hard disk.In our test cases，the de⁃fault local cache size of a cloud disk client is 100MB;the test file size is 50 MB;and the data block size of IO request in⁃creases from 32 kB to8192 kB in doubling size.

▲Figure4.Cache replacement timesof fileoperations.

The throughput is shown in Figs.5 and 6.The blue line represents the write⁃read speed of the cloud disk and the red line represents the write⁃read speed of the local disk.The speed ofaccessing data to the cloud disk is slightly slower than the local disk due to the extra processing of the file data.How⁃ever，from the user’s perspective，there is no difference be⁃tween the two read⁃write speedswhen the usermanipulateshis/ her data from the disk device.The cloud disk performswell in the read⁃write throughput for users.

5 Conclusion

In this paper，we have presented a virtual disk mapping method for cloud desktop storage.Unlike state⁃of⁃the art pro⁃posals of the cloud disk，we use virtual disk driver and file sys⁃tem filter driver to simulate a virtual desktop disk associate with users’cloud storage space.The virtual disk driver pro⁃vides a user⁃friendlymode for accessing desktop data.It also provides a flexible cache spacemanagementmechanism.The file system filter driver intelligently checks I/O requests of up⁃perapplicationsand synchronizes file accessing requests tous⁃ers’cloud storage services.The evaluation results show that the read⁃write performance of our virtual diskmappingmethod with customizable local cache storage is almostsame as thatof the localhard disk.

▲Figure5.Read throughput.

▲Figure6.W rite throughput.

In the futurework，the data encryption of the clientbased on the file system filter driverwill be studied and the I/O per⁃formance of cloud disk willbe continually optimized.

[1]J.Wu，L.Ping，X.Ge，Y.Wang，and J.Fu，“Cloud storage as the infrastructure of cloud computing，”2010 Int.Conf.Intelligent Computi.Cognitive Informatics, Kuala Lumpur，Malaysia，pp.380-383.doi:10.1109/ICICCI.2010.119.

[2]W.Zeng，Y.Zhao，K.Ou，and W.Song，“Research on cloud storage architecture and key technologies，”in Proc.2nd Int.Conf.Interaction Sci Inform.Technol., Culture and Human，Seoul，Korea，Nov.2009，pp.1044-1048.doi:10.1145/ 1655925.1656114.

[3]K.Beaty，A.Kochut，and H.Shaikh，“Desktop to cloud transformation plan⁃ning.”in IEEE Int.Symp.Parallel&Distributed Process.,Rome，Italy，May 2009，pp.1-8.doi:10.1109/IPDPS.2009.5161236.

[4]Citrix XenDesktop[Online].Available:http://www.citrix.com/products/xendesk⁃top/overview.html

[5]VMware.(2012).VMware view 4.5:modernize desktop and applicationmanage⁃ment，v.2.0，brochure.[Online].Available:http://www.vmware.com/files/pdf/VM⁃ware⁃View⁃45⁃DS⁃EN.pdf

[6]I.Drago，M.Mellia，M.M.Munafo，A.Sperotto，R.Sadre，and A.Pras，“Inside dropbox:understanding personal cloud storage services，”in Proc.2012 ACMConf.Internet Measurement，Boston，USA，pp.481-494.doi:10.1145/ 2398776.2398827.

[7]GoogleDrive[Online].Available:https://www.google.com/drive/

[8]Baidu CloudDisk[Online].Available:http://yun.baidu.com/

[9]Dropbox[Online].Available:https://www.dropbox.com/

[10]Kingsoft Kuaipan[Online].Available:http://www.kuaipan.cn/

[11]T.Chen，K.H.Lee，Y.Sakane，and T.T.Yamashita，“Hard disk drive system having virtual contact recording，”U.S.Patent5 673 156，Sept.30，1997.

[12]K.Kerr.(2011，Feb.23).The virtual disk APIin Windows 7.MSDNMag.[On⁃line].Available:http://msdn.microsoft.com/en⁃us/magazine/dd569754.aspx?pr= flas1

[13]A.Schulman，D.Maxey，and M.Pietrek，Undocumented Windows:A Program⁃mer’sGuide to Reserved MIicrosoftWindowsAPIFunctions，Boston，USA:Addi⁃son⁃Wesley，1992.

[14]P.E.Soran，J.P.Guider，L.E.Aszmann，and M.J.Klemm，“Virtual disk drive system andmethod，”U.S.Patent7 613 945，Nov 3，2009.

[15]S.R.McDowell，“File system filter driver apparatus andmethod，”U.S.Patent 6 266 785，Jul.24，2001.

[16]R.Nagar，WindowsNT File System Internals:ADeveloper’sGuide.Sebastopol，USA:O’Reilly Media，1997.

[17]L.Zheng，Z.Ma，and M.Gu，“Techniques of file system filter driver⁃based and security⁃enhanced encryption system，”J.Chinese Comput.Syst.，vol.28，no.7，pp.1181-1184，Jul.2007.doi:10.3969/j.issn.1000⁃1220.2007.07.006.

[18]Iozone Filesystem Benchmark[Online].Available:http://www.iozone.org/

[19]W.Vogels，“File system usage in Windows NT 4.0，”ACMSIGOPSOperating Syst.Rev.，vol.34，no.2，pp.17-18，Apr.2000.doi:10.1145/346152.346177.

Biographies phies

Hancong Duan(duanhancong@163.com)received his BS degree in computer sci⁃ence from Southwest Jiaotong University in 1995，ME degree in computer architec⁃ture in 2005，and PhD degree in computer system architecture from University of Electronic Science and Technology of China in 2007.He is currently a professor of computer science at the University of Electronic Science and Technology of China. His current research interests include large⁃scale P2Pcontentdelivery network，dis⁃tributed storage，and operating system.

Xiaoqin W ang(xiaoqinwang0508@gmail.com)received his BS degree in computer science from University of Electronic Science and Technology of China.She is cur⁃rently a ME candidate at the Departmentof Computer Software and Theory，Univer⁃sity of Electronic Science and Technology of China.Her research interests include distributed storageand cloud computing.

Ping Lu(lu.ping@zte.com.cn)received his ME degree in automatic control theory and applications from South EastUniversity.He is the chiefexecutiveof the Service Institute of ZTE Corporation.His research interests include augmented reality and multimedia services technologies.

Shengmei Luo(luo.shengmei@zte.com.cn)received his MS degree in telecommuni⁃cation and electronics from Harbin Institute of Technology in 1996.He is a chiefar⁃chitect at ZTE corporation.His research interests include cloud computing，cloud storage，and big data.

ZhiyongW ang(wang.zhiyong@zte.com.cn)received hismaster’sdegree in earth ex⁃ploration and information technology from China University of Mining.He is cur⁃rently a product planningmanager at ZTE Corporation.His research interests in⁃clude cloud computing，cloud storage，and big data.

t

2014⁃06⁃16

10.3969/j.issn.1673-5188.2014.04.001

http://www.cnki.net/kcm s/detail/34.1294.TN.20141216.1723.001.htm l,published on line December 16,2014

This wo rk on key techno logies o f the in teg ration of cloud desktop and

cloud sto rage Platfo rm is suppo rted by ZTE Industry⁃Academ ia⁃Research Cooperation Funds.