2. DISTRIBUTED CACHE SYSTEM EE SOFTWARE LAB, TECHNION By Shamil Nisimov Dror Bohrer Supervisor : Yaron Ben Shoshan Lab Engineer : David Ilana

3. Distributed Cache System2 Introduction Client wants a shared mean of storage on the network – A Server that will use as storage and maintenance of information. There is a need for performance and scalability qualities in a protocol that will allow duplicating and synchronizing servers in the network.

4. Distributed Cache System3 Introduction - cont The system will be protected from failure of servers and network links, will detect changes of information in the network and will do all the synchronization necessary. The System will enhance its performance and reduce the traffic by updating servers only on demand.

5. Distributed Cache System4 Key Terms  Cell – exist in each server contains information, in this project – textual string, can be HTML page.  Version – the sequence number of given data stored in a cell. Higher Version = Newer data

6. Distributed Cache System5 iBus Middleware In this project we used iBus to connect between servers by group communication. iBus provides the following key qualities : Reliable – no need for resend. Ordering – between network packets. View Change – every server fall is detected by the bus and reported to other servers.

7. Distributed Cache System6 Project Model All servers has the same cells but each cell can contain different local version In every server Server 2 Client iBus Client Server 3 Server 0 Server 1

8. Distributed Cache System7 Distributed System Pros  server holds data only on demand (optimizing traffic in the network).  Service is faster, because clients can connects to their closest server  Reliable – Due to iBus middleware.  Scalable – A server can be added/removed (partition).

9. Distributed Cache System8 Distributed System Cons  Data is stored in many servers instead of in one permanent server.  There is more traffic overhead when servers update each other.

10. Distributed Cache System9 Client Capabilities  can connect to closest server  Local request  Global request  Client Update Data

11. Distributed Cache System10 1. Local Request  the client requests from the server the local version exist in the server (if bigger then the version holds by the client).  Advantage – faster service.  Disadvantage – older information.

12. Distributed Cache System11 2. Global Request  the client requests the server the most up to date version exists in the network.  Advantage – The most up to date data in the network.  Disadvantage – slower service.

13. Distributed Cache System12 Server 2 Client iBus Client Server 3 Server 0 Server 1 Local/Global Request-example On cell no.0 client issue local request when his version is 2 Server has Version 3 so he send client this version. On cell no. 0 client issue global request when his version is 2 Server has Version 1 global version not valid so he can not respond the client. Issues global request to all servers Server 1 responds – my version is 1 Server 2 responds – my version is 1 Server 3 responds – my version is 3 Server 0 request server ’ s 3 related data Server 3 sends its data related to version 3 Server updates the client with version 3

14. Distributed Cache System13 3. Client Update Data the client updates his server and by that making its version the most up to date in the network.

15. Distributed Cache System14 Server 2 Client iBus Client Server 3 Server 0 Server 1 Client update data - example Client updates his server on cell 0 Server 3 -Global Version is valid (version 3) server opens transaction Send phase1 packet on cell 0 to all servers Now All Server finds this packet on phases queue. This move them to phase two state and locker is server 3. Server sends phase2 packet on cell 0, Saying: new global version - 4 Send back response to client meaning: update was successful, new version - 4

16. Distributed Cache System15 Server Structure – External  Contains cells each represent different subject (HTML page).  For each cell holds to version related to data in server.  Holds cell global version (if known to him).

17. Distributed Cache System16 Server Structure - Internal iBus middleware SERVER BUS CONNECTION RECEIVE QUEUE FIFO THREAD SERVER CELLS (CLASS) TRANSMIT QUEUE FIFO THREAD GUI REQUEST HANDLING THREADS TRANS- ACTIONS HANDLING THREADS IBus Connection of server to iBus through which packets are sent Here packet are sorted according to packet opcode Packets wait to send here in a fifo matter Here all version calculation are made Here implemented transaction algorithm All cell relevant information is stored here

18. Distributed Cache System17 Server 1 Client iBus Client Server 2 Simulation Stub Server 0 Project Simulation The stub here can choose which server belong to which group and by that create partition in the network. In order to check system behaviour on cases like partition and Server falls we created a simulation stub.

19. Distributed Cache System18 Possible future expansion Exchange of HTML pages Supporting of HTTP protocols. TO BE CONTINUED ….