Skip to content

Latest commit

 

History

History
156 lines (127 loc) · 4.72 KB

README.md

File metadata and controls

156 lines (127 loc) · 4.72 KB

CS-739-P3: Replicated Block Store

Replication Strategy

  1. We use Primary-Backup. We support only 2 server nodes. Source code: src/blockstorage_server.cc
  • class BlockStorageServiceImpl: responds to client requests of Read(Addr) and Write(Addr)
  • class ServiceCommImpl: communicates with the other server node to propogate writes
  • class ServiceCommClient: communicates witht the other server for recovery
  • class LBNodeCommClient: communicates with the Load Balancer to send heartbeats and identify failover

  1. Client library Souce code: src/client.h and src/client/cc

  2. Load Balancer Source code: src/load_balancer.cc

  • class BlockStorageService: forwards client request to a server
  • class LBNodeCommService: used to identify server unavailability

Durability

We use a two phase strategy of writing to a temp file and atomic rename from temp to orginal file
Source code: src/blockstorage_server.cc

Crash Recovery Protocol

We perform 4 steps to recover from a server crash viz.

  1. Parse log file
  2. Get pending writes
  3. Recover from other states
  4. Cleanup
    Source code: src/util/crash_recovery.h and src/util/crash_recovery.cc

Failover

Servers exchange hearbeats with the load balancer
Source code: src/blockstorage_server.cc and src/load_balancer.cc

Test code

Present in performance/ and MACROS present in src/util/common.h

  • concurrent_rw.cc: Client code that issues reads and writes concurrently
  • latency_perf.cc: Client code that issues the read and write multiple times on the same address
  • recovery_perf.cc: Client code that initialises the servers with writes to analyse recovery time
  • rw_perf.cc: Client code that issues reads and writes on multiple addresses concurrently

Utilities

  1. Address Translation Source code: src/util/address_translation.h and src/util/address_translation.cc

  2. Cache Source code: src/util/cache.h and src/util/cache.cc

  3. Key Value Store Source code: src/util/kv_store.h and src/util/kv_store.cc

  4. Reader-Writer Locks Source code: src/util/locks.h and src/util/locks.cc

  5. Handling in-flight and future writes during recovery Source code: src/util/txn.h and src/util/txn.cc

  6. Write Ahead Logging Source code: src/util/wal.h and src/util/wal.cc

Setup

gRPC Installation

Follow these steps to install gRPC lib using cmake: https://grpc.io/docs/languages/cpp/quickstart/#setup. :warning: make sure to limit the processes by passing number(e.g. 4) during make -j command.

for example, instead of make -j use make -j 4

Build

Main source code

  1. cd src/
  2. chmod 755 build.sh
  3. ./build.sh

Performance scripts

  1. cd performance/
  2. chmod 755 build.sh
  3. ./build.sh

Run

Client

cd src/cmake/build
./blockstorage_client

OR

cd performance/
./build
cd cmake/build
<ANY OF THE EXECUTABLES LISTED IN THIS DIRECTORY>

Load Balancer

./load_balancer

Primary Server

./blockstorage_server PRIMARY [self_addr_lb] [self_addr_peer] [peer_addr] [lb_addr]

Eg.

./blockstorage_server PRIMARY 20.127.48.216:40051 0.0.0.0:60052 20.127.55.97:60053 20.228.235.42:50056

Backup Server

./blockstorage_server BACKUP [self_addr_lb] [self_addr_peer] [peer_addr] [lb_addr]

Eg.

./blockstorage_server BACKUP 20.127.55.97:40051 0.0.0.0:60053 20.127.48.216:60052 20.228.235.42:50056

Performance Scripts

  • concurrent_rw
cd performance/cmake/build
./concurrent_rw
  • latency_perf
cd performance/cmake/build
./latency_perf -a <num> -j <num> -l <num> -i <num> -t <num>

Where a stands for the start address j stands for the next address jump l stands for the count of addresses / limit i stands for the number of iterations t stands for the test, 0 for read and 1 for write

  • recovery_perf
./recovery_perf -a <num> -j <num> -l <num> -i <num>

Where a stands for the start address j stands for the next address jump l stands for the count of addresses / limit i stands for the number of iterations

  • rw_perf
./rw_perf <num> <num>

Where the first num corresponds to 0 for read and 1 for write. The second num corresponds to the number of worker threads.

Deliverables

  1. Demos
  2. Report
  3. Presentation