SETUP

== Dependencies == nanomsg (nanomsg.org) jemalloc (www.canonware.com/jemalloc)

HOW TO RUN

Create a file called ifconfig.txt with IP addresses for the servers and clients, one per line.

$ make $ ./rundb -nid[unique server ID] & $ ./runcl -nid[unique client ID] &

DBMS BENCHMARK

== General Features ==

ddbms is a OLTP database benchmark with the following features.

1. Seven different concurrency control algorithms are supported. NO_WAIT[1] : no wait two phase locking WAIT_DIE[1] : wait and die two phase locking TIMESTAMP[1] : basic T/O MVCC[2] : multi-version T/O OCC[3] : optimistic concurrency control (MaaT) CALVIN[4] : Calvin

[1] Phlip Bernstein, Nathan Goodman, "Concurrency Control in Distributed Database Systems", Computing Surveys, June 1981 [2] Phlip Bernstein, Nathan Goodman, "Multiversion Concurrency Control -- Theory and Algorithms", Transactions on Database Systems, Dec 1983 [3] H. Mahmoud et al, "MaaT: Effective and Scalable Coordination of Distributed Transactions in the Cloud", VLDB 2014 [4] A. Thomson et al, "Calvin: Fast Distributed Transactions for Partitioned Database Systems", SIGMOD 2012

2. Three benchmarks are supported. 2.1 YCSB[5] 2.2 TPCC[6] Only Payment and New Order transactions are modeled. 2.2 Product-Parts-Supplier (PPS)

[5] B. Cooper et al, "Benchmarking Cloud Serving Systems with YCSB", SoCC 201 [6] http://www.tpc.org/tpcc/

== Config File ==

dbms benchmark has the following parameters in the config file. Parameters with a * sign should not be changed.

• CORE_CNT : number of cores modeled in the system.

• NODE_CNT : number of computation nodes modeled in the system

• PART_CNT : number of logical partitions in the system; Should equal CORE_CNT * NODE_CNT

• THREAD_CNT : number of worker threads

• REM_THREAD_CNT : number of message receiver threads

• SEND_THREAD_CNT : number of message sender threads

• TPORT_TYPE : communication protocol (TCP, IPC)

• ENVIRONMENT_EC2 : true if running on Amazon EC2

• PAGE_SIZE : memory page size

• CL_SIZE : cache line size

• DONE_TIMER : number of nanoseconds to run experiment

• WARMUP_TIMER : number of nanoseconds to run for warmup

• SEQ_BATCH_TIMER : number of nanoseconds between CALVIN batches

• WORKLOAD : workload supported (TPCC, YCSB, or PPS)

• CC_ALG : concurrency control algorithm (WAIT_DIE, NO_WAIT, TIMESTAMP, MVCC, CALVIN, or MAAT)

• ISOLATION_LEVEL : system isolation level (SERIALIZABLE, READ_COMMITTED, READ_UNCOMMITTED, or NOLOCK)

• THREAD_ALLOC : per thread allocator.

• * MEM_PAD : enable memory padding to avoid false sharing.

• MEM_ALLIGN : allocated blocks are alligned to MEM_ALLIGN bytes

• * ROLL_BACK : roll back the modifications if a transaction aborts.

• ENABLE_LATCH : enable latching in btree index

• * CENTRAL_INDEX : centralized index structure

• * CENTRAL_MANAGER : centralized lock/timestamp manager

• INDEX_STRCT : data structure for index.

• BTREE_ORDER : fanout of each B-tree node

• TS_TWR : enable Thomas Write Rule (TWR) in TIMESTAMP

• HIS_RECYCLE_LEN : in MVCC, history will be recycled if they are too long.

• MAX_WRITE_SET : the max size of a write set in OCC.

• MAX_ROW_PER_TXN : max number of rows touched per transaction.

• QUERY_INTVL : the rate at which database queries come

• MAX_TXN_PER_PART : maximum transactions to run per partition.

// for YCSB Benchmark

• SYNTH_TABLE_SIZE : table size
• ZIPF_THETA : theta in zipfian distribution (rows accessed follow zipfian distribution)
• READ_PERC :
• WRITE_PERC :
• SCAN_PERC : percentage of read/write/scan queries. they should add up to 1.
• SCAN_LEN : number of rows touched per scan query.
• PART_PER_TXN : number of logical partitions to touch per transaction
• PERC_MULTI_PART : percentage of multi-partition transactions
• REQ_PER_QUERY : number of queries per transaction FIRST_PART_LOCAL : with this being true, the first touched partition is always the local partition.

// for TPCC Benchmark

• NUM_WH : number of warehouses being modeled.
• PERC_PAYMENT : percentage of payment transactions.
• DIST_PER_WH : number of districts in one warehouse
• MAX_ITEMS : number of items modeled.
• CUST_PER_DIST : number of customers per district
• ORD_PER_DIST : number of orders per district
• FIRSTNAME_LEN : length of first name
• MIDDLE_LEN : length of middle name
• LASTNAME_LEN : length of last name

// for PPS Benchmark

• MAX_PPS_PART_KEY : number of parts in the parts table
• MAX_PPS_PRODUCT_KEY : number of products in the products table
• MAX_PPS_SUPPLIER_KEY : number of suppliers in the suppliers table
• MAX_PPS_PARTS_PER : number of parts per product or supplier