Mobile gpu cloud computing

Editor's Notes

• #5: Mobile cloud computing (MCC) has been introduced to be a potential technology for mobile services. MCC integrates the cloud computing into the mobile environment overcomes obstacles related to the performance (e.g., battery life,storage, and bandwidth), environment (e.g., heterogeneity, scalability, and availability), security (e.g., reliability and privacy).
• #6: Efficient and effective computation offloading: Variable bandwidth Mobile service availability Difficulty of runtime conditions estimation Security: Multi-tenancy in cloud causes vulnerability Offloaded code prone to tampering End-to-end security of mobile code at risk
• #11: Cloud Computing is a combination of virtualization (one computer hosting several “virtual” servers), automated provisioning (servers have software installed automatically) distributed computing and Internet connectivity technologies to provide the service.
• #12: Iaas:Provides infrastructure such as computer instances, network connections, and storage so that user can run any software or operating system. Paas:Provides the ability to deploy applications through a programming language or tools supported by the cloud platform provider. Saas:Provides the ability to use the software in a cloud environment
• #13: Private Cloud which the services and infrastructure are maintained on a private network. Public Cloud which the services and infrastructure are provided off-site over the Internet.  Hybrid Cloud which is composition of both public and private clouds.
• #14: Reasons for Building a Private Cloud The better utilization of hardware The ability to quickly provi­sion capabilities The ability to easily provide continuity of application. B. Categories of Infrastructure for a Private Cloud Before Selection the private cloud infrastructure software we need to understand the component of private cloud infrastructure • Virtualization: technology provides the hypervisor software, which allows multiple instances of “guest” operating systems to run concurrently on the same physical server. • Storage technology: such as storage area networks (SANs), can provide the underlying persistent storage for the data center. • Security such as identity management and virtual firewalls. • Provisioning, management, and metering tools provide the ability to deploy virtual operating systems, services, and applications.
• #15: The reason of selection openStack: a) Provide scalable, elastic cloud computing for private cloud. b) Simple to implement and massively scalable Change. c) Managed all resources through a dashboard that gives administrators control while empowering their users to provision resources through a web interface. d) Excellent for large deployments. e) The user can have good knowledge to the used component only not all openstack components. 2) OpenStack Component And Architecture : • Dashboard ("Horizon"): provides a modular web based user interface for all the OpenStack services. • Identity ("Keystone"): provides authentication and authorization • Image ("Glance") : provides a catalog and repository for virtual disk images. • Compute ("Nova") : provides virtual servers upon demand. • Networking ("Neutron"): Enables network connectivity as a service among interface devices managed by other OpenStack services, usually Compute Service. • Block Storage ("Cinder "): A service that manages persistent block storage volumes for virtual machines. • Object Storage ("Swift "): provides object storage. It allows you to store or retrieve files.
• #16: The reason of selection OpenShift. Open Source Software and run on top of Openstack private cloud infrastructure Openshift provide programming languages and frameworks Java, Ruby, and PHP. It also provides integrated developer tools to support the application lifecycle,including Eclipse integration, JBoss Developer Studio. Easy to deploy and provide great documentation for the services. Provide standard languages , IDE Integration , Built-in Database Services and responsive Web Console and application portability.
• #17: This web application is RESTful webservice return result in JSON format.   VM One Server Multi Server IaaS 3.60 sec 1.75 sec 1.02 sec PaaS 5.95 sec 2.83 sec 1.98 sec
• #19: In this architecture The cloud resource is placed in a remote centralized cloud infrastructure. The mobile devices access the backbone network via WiFi access points (APs) or 3G/4G cellular networks. The Requests and information of Mobile users are transmitted to the central processors that are connected to servers providing mobile network services. After that, the requests are delivered to a cloud through the Internet. In the cloud, cloud controllers process the requests to provide mobile users with the corresponding cloud services .
• #20: The mobile interact with cloud service by two ways: Native Mobile applications(used in our application): are developed using mobile platform supported programming languages and a set of APIs provided by the cloud service provider . This type is used in the face recognition application as android application is installed in the mobile device to allow to send request to the face recognition cloud engine. Embedded browser applications: are developed using standard web development languages (e.g. HTML and JavaScript). Web service : Send data as json object not xml
• #25: first part is Mobile application (Android platform) and second part is face recognition engine on the cloud. The mobile communicate with the cloud by using the restful web service as the image and user data is send to the web server in JSON format. The user data is used to authorize the application users. Also the image is encrypted in before send it. And decrypted once arrive to the cloud engine. The First part is face detection :is based on identifying and locating a human face in images regardless of size, position, and condition The Second part is face recognition: it is analysis part as it compare the detected person face with the stored human faces in database. The face recognition mobile cloud application is native application which has to deploy in mobile. The face recognition cloud engine it apply the face detection algorithm Viola Jones Face Detector at the first Then for the detected faces it converts them into a set of Eigenfaces and compare with stored faces If the identification percentage 80% or more the response will back to client mobile application with the person information else it will inform the client that face is unknown.
• #26: AdaBoost و Integral Image فيولا -- ادابووست The face detection algorithm looks for specific Haar features of a human face. When one of these features is found the algorithm allows the face candidate to pass to the next stage of detection. A face candidate is a rectangular section of the original image called a sub-window. This sub-window is often scaled in order to obtain a variety of different size faces. The Viola and Jones face detection algorithm uses a cascade of stages which is used to eliminate non-face candidates quickly. The cascade eliminates candidates by making stricter requirements in each stage with later stages being much more difficult for a candidate to pass. Candidates exit the cascade if they pass all stages or fail any stage. A face is detected if a candidate passes all stages. Each stage consists of many different Haar features.
• #27: The face recognition is done via a series of steps. Eigenfaces Initialization The set of faces are grouped together into a training set. Image database consist of 50 face image for 10 unique people with 5 photos to each person in different lighting, expressions and backgrounds The mean of the training set is found and subtracted from the training set. The covariance matrix is calculated. From the covariance matrix eigenvectors and the eigenvalues for each vector are calculated. Eigenfaces recognition the best eigenfaces are selected, which are the ones whose eigenvalues are the highest. Compare input image with our mean image and multiply their difference with each eigenvector of the L matrix. Each value would represent a weight and would be saved on a vector Ω
• #29: Image processing: Image database consist of 50 photo to 10 persons. Internet Speed: Average speed Wi-Fi network (). Mobile 3G networking. Application Assumptions: Resend the request if mobile device doesn’t get the response within 2 min
• #34: CPU : powerful processor and big caches with few number of cores , small memory bandwidth (21 GB /sec) GPU: small processors and small cash with large number of cores (up to 1000) , huge memory bandwidth 177 GB /sec
• #35: 15 * 32 *48 =23040 Graphics have a lot of parallism pixels polygons
• #39: GPU Implementation Here, we will briefly see CUDA implementation of Viola Jones from "Towards a Robust, Real-time Face Processing System using CUDA-enabled GPUs" by Sharma et al. They have mainly targeted nVIDIA GeForce GTX 285 GPU. This GPU features 30 multiprocessors, 16 KB shared memory per multiprocessor and 1 GB device memory. There can be a maximum of 512 threads per block and 1024 active threads per multiprocessor. As described previously, face detection mainly consists of, calculating the integral images for fast feature evaluation, and detecting faces using a cascade of classifiers. Each of these tasks are parallelized and run as kernels on the GPU.
• #40: Experiment Goal: measure the execution of face recognition application on local on mobile device and remotely on deferent cloud platform Experiment description: The images database consists of 50 photos to 10 persons. The three test images will be tested the first image contains one face to detect and recognition the second image contains three faces and the third image with seven faces. Each face will be execute on face recognition mobile application, face recognition with virtual machine cloud engine, face recognition with one server cloud engine and multiple server cloud engine
• #45: Registration: each mobile must register to the cloud with unique identifier and some other security parameters. Generate encryption key: mobile and cloud generate the same encryption master key (EMK) depend in the mobile id and security parameters. Cloud message: the cloud reply to user with authorization message which is consist of two parts first register request status (accept/reject) if request had been accepted the second part will be cloud decryption key (CDK). Divided data and encryption process: the mobile will divide the transferred data to small parts according to part size in security parameter. Each part will be encrypted with (EMK + part number) and all encrypted parts with integrated again with the same order and reencrypt the package with (EMK).Mobile send encrypted data to cloud and The cloud will decrypt the data and apply the required processing. The cloud will encrypt the result with (CDK). Result decryption: mobile will decrypt the received result by using (CDK). The registration will be one time only , and each time check authentication and authorization Cloud store EMK /CDK for each mobile. Mobile also store EMK /CDK