What if We Atomize Student Data and Apps and Put Them on Docker Containers?
1 like88 views
The document proposes atomizing student data and applications into Docker containers to create a hybrid cloud campus environment. Key points include: - Using multiple containers per student instead of individual VMs for better resource utilization and mobility across campus infrastructure. - Student web applications and data can remain unchanged and be accessed via container IDs instead of direct hardware. - Docker provides benefits like layered builds to simplify deployments across heterogeneous hardware like PCs and Raspberry Pis. - Performance tests show Docker containers have comparable or better performance than VMs, though Raspberry Pis have limitations due to their hardware.
What if We Atomize Student Data and Apps and Put Them on Docker Containers?
- 1. Marat Zhanikeev maratishe@gmail.com maratishe.github.io and Put Them on Docker Containers? Tokyo Univ. of Science We Atomize Student Data and Apps ICM Cloud Academy (ICACON) @ Tokyo PDF → bit.do/180524 #cloudification #VMs #containers #campus #university #offload #fogclouds #IoT #raspberrypi #docker What if
- 2. 2 Ways to Deal with Network Congestion To 3G/LTE Virtual Wireless User Internal Engine To 3G/LTE Resource Virtualization 1. use multiconnect to pool and virtualize group connectivity as in the figure 05 2. as in this paper, offload network activity 05 myself ”Virtual Wireless User: A Practical Design for Parallel MultiConnect Using WiFi Direct in Group Communication” 10th MobiQuitous (2013) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 2/15 2/15
- 3. Traditional vs Cloudified Campus • since VM-per-student is not feasible, multiple-containers-per-student is the base unit • containers can easily migrate across hardware within the campus (assuming virtualized hardware) • can be viewed as a kind of hybrid cloud Access Digital campus Traditional University Students Professors Access from home Local access Cloudified University Hybrid Cloud Access from home Virtual machine Containers Physical machine Container on PM Unmanaged hardware Local access Virtual classroom M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 3/15 3/15
- 4. Data and Apps on Containers • huge pro: no need to change anything within the webapp (even keep the id-based authentication) • a small con: students now have to lookup their individual webapp containers Traditional webapps Data (files, etc.) SQL Database Web app User (student, lecturer, etc) Web Files SQL DB Web Portal Where is my container? (User ID) Hybrid Cloud Virtual user (web + data + SQL + … containers) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 4/15 4/15
- 5. Implementation (Docker) • hardware should be virtualized – Xen and such – for better management in heterogeneous environments • RPi is special – no way to run VMs, so only container level is possible • otherwise, Docker is the best tool on the market today ◦ recent innovation of layered builds drastically simplifies cloning and reduces deploy time Raspberry Pi Xen PM Linux Xen Virtualizer Docker Docker App Docker App … Linux VM … Docker Docker App … M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 5/15 5/15
- 6. Webapp performance (docker vs...) 1 2 5 10 20 50 100 200 500 1000 2000 0 2000 4000 6000 8000 10000 12000 14000 Averagedownloadtime(ms) Linux VMWindows VMLinux BoxDocker container on RPi 1 2 5 10 20 50 100 200 500 1000 2000 0 100 200 300 400 500 600 700 800 Averagedownloadtime(ms) Downloaded data size (kbytes) • up to 2Mbyte replies to webapp requests • Docker on RPi stands out (in a bad way) – is it Docker or is it RPi? M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 6/15 6/15
- 7. Webapp performance (closer look) • obviously, bad performance is due to RPi, not Docker RPi noDocker (direct webapp) Docker on virtual CentOS RPi Docker on LAN 1 2 5 10 20 50 100 200 500 1000 2000 0 2000 4000 6000 8000 10000 12000 14000 Averagedownloadtime(ms) Downloaded data size (kbytes) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 7/15 7/15
- 8. Deployment performance • deploy using layered build – that is, having the majority of dependencies present locally • RPi does not just take longer, but the deploy time is much more unpredictable • otherwise, even RPi can accommodate dozens of containers (freezes sometimes, though) 1 3 5 10 20 30 50 2 6 10 14 18 22 Taskcompletiontime(s) Adding containers Removing containers Size of container population 1 3 5 10 1 3 5 7 9 11 Taskcompletiontime(s) Adding containers Removing containers Size of container population M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 8/15 8/15
- 9. Docker Volumes (are just paths) • data migration is trivial – just rsync it ... ◦ between master and slave copies ◦ between locations • binary diffs 08 can help further increase efficiency 08 myself ”DiffHub: An Efficient Cloud Sync Technology based on Binary Diffs” IPSJ JIP, vol.23, no.5 (2015) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 9/15 9/15
- 10. Future : Campus as a Fog Cloud • push vs pull deploys, orchestration, and local resource discovery 01 • fox boxes without internet connection in the campus 07 WiFi Wireless users WiFi AP Physical Device Cloud Platform VM VM Con.Con.Con. Storage Sensors … Beacon WiFi AP WiFi Client P2P WiFi Box s Services A Fog Box VM VM VM … Pull Deploy Device Provider 3rd Party Cloud Fog Local Services Local wireless users P2P WiFi P2P WiFi-only networking 01 myself ”A Cloud Visitation Platform to Facilitate Cloud Federation and Fog Computing” IEEE Computer (2015) 07 myself ”Theory and Practice for Fog Infrastructure based on Standalone Cloudified IoT Boxes” IEEE TRON Symposium (2017) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 10/15 10/15
- 11. That’s all, thank you ... M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 11/15 11/15
- 12. Total Cloudification WiFi Scale/sizeofcloudifieditems Timeline Now (2016) Probably should not cloudify Should definitely cloudify Data Centers Racks IoV > VC Desktop Notebook APs, Beacons Smartphone Sensor IoT Delegated networking Network core Network edge 07 myself ”Theory and Practice for Fog Infrastructure based on Standalone Cloudified IoT Boxes” IEEE TRON Symposium (2017) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 12/15 12/15
- 13. Fog Box with Local Resource Discovery • storage is the first obvious choice for local resource 02 • but sensors, connectivity, etc. are also valid resources • LHAP : Local Hardware Awareness Platform 01 is also the obvious answer to the heterogeneity problem Physical Device LHAP VM VM Con.Con.Con. HUB WiFi Wireless users Wired users Wireless AP Storage Sensors … WiFi Wireless users WiFi AP Physical Device Cloud Platform VM VM Con.Con.Con. Storage Sensors … Beacon WiFi AP WiFi Client P2P WiFi Box s 02 myself ”Fog Caching and a Trace-Based Analysis of its Offload Effect” IJITSA, vol.10(2) (2016) 01 myself ”A Cloud Visitation Platform to Facilitate Cloud Federation and Fog Computing” IEEE Computer (2015) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 13/15 13/15
- 14. Cloud Boxes : Push vs Pull Deploy • pull is best for closed, tightly controlled systems with high security (set-top boxes?) • push is necessary for large-scale cloud and fog federations Cloud Visitation Platform App App App … Pull-Based Installation (of Apps) End User Device Provider3rd Party Service Local services One Cloud Cloud Visitation Platform App App App … Cloud/device Owner 3rd Party Service Federated Clouds Local services Federated Cloud Manager 01 myself ”A Cloud Visitation Platform to Facilitate Cloud Federation and Fog Computing” IEEE Computer (2015) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 14/15 14/15
- 15. Local Hardware Awareness Platform Local hardware VirtualMachine Cloud Platform (Xen) App App App … Multitenant environment Physical Machine (PM) Tools API Layer Control Layer Local Hardware Awareness Platform (LHAP) … VirtualMachine App Tools … VirtualMachine App VirtualMachine App App App … Global network (to Internet) In-box network Conventional Platform … MiniOS App SSD Operating System VirtualMachine App App App … Multitenancy impossible Physical Machine (PM) Cloud Platform (Xen) Operating System Global network (to Internet) HDD Local Network Global network (to Internet) Physical Machine (PM) Cloud Platform (Xen) Operating System DtL: Direct to Local Local hardware SSD HDD Mini Cache LHAP (proposed) M.Zhanikeev – maratishe@gmail.com What if We Atomize Student Data and Apps and Put Them on Docker Containers? – bit.do/180524 15/15 15/15