Don’t Lose Sleep Over Availability The GreenUp Decentralized Wakeup Service Siddhartha Sen, Jacob R. Lorch, Richard Hughes, Carlos Garcia Jurado Suarez, Brian Zill, Weverton Cordeiro, and Jitendra Padhy Presentation: Kamil Yurtsever
Context ● We have many machines grouped in relatively small subnetworks. ● These are typical office machines. http://en.wikipedia.org/wiki/File:Computer-aj_aj_ashton_01.svg
Problem We want ● machines to be ○ available for remote access all the time ○ asleep as much as possible ● solution to be ○ easy to deploy ○ cheap ○ nonintrusive http://commons.wikimedia.org/wiki/File:USCurrency_Federal_Reserve.jpg
What we can't have ● virtualization ● special hardware ● dedicated machines http://commons.wikimedia.org/wiki/File:Stop.png
General Idea ● let machines sleep ● awake machines listen if there are packets meant for sleeping machines and wake them if necessary http://commons.wikimedia.org/wiki/File:2010-07-20_Black_windup_alarm_clock_face_SVG.svg
Ingredients ● Wake-on-LAN ● Ethernet port hijacking ● Distributed proxies http://en.wikipedia.org/wiki/File:Achiote_paste_ingredients.jpg
Assumptions ● Subnet domains ● Availability over consistency ● Load-sensitive, unreliable machines
Key terms ● participant ● manager ● managee ● state http://en.wikipedia.org/wiki/File:Crypto_key.svg ● proxy ● distributed management ● subnet state coordination ● apocalypse ● guardians
Distributed management ● Distributed probing ● Multiple management problem ● Manager duties
Subnet state coordination ● Periodic broadcast ● Rebroadcast by managers ● Roll call
Preventing apocalypse ● Non-independent sleep behavior ● Guardians http://commons.wikimedia.org/wiki/File:Asteroid_1.png
Improvements ● dynamic layer 2 routing ● handling encrypted traffic ● increasing availability ● choosing better guardians
The End Thank you for your attention.
Recommend
More recommend