@andy_pavl o OLTP on NVM: YM MV

The Last Six Months ? PDL Retreat October 2013 PDL Visit Day May 2014

Prison Life Washing Dishes Not Fighting Repentant Cafeteria Thievery Shankings Making Pruno G O OD EV IL

NVM OLTP Lightweight CC Logical Logging Snapshots Heavyweight CC ARIES Logging Making Pruno DR AM SSD/ HDD

Overview 9 Understand the performance characteristics of NVM to develop an optimal DBMS architecture for OLTP workloads.

OLTP Workloads 10 Transactions have three defining characteristics: –Short-lived. –Small footprint. –Repetitive.

Intel NVM Emulator 11 Instrumented motherboard that slows down access to the memory controller. Two execution interfaces: –NUMA (NVM-only) –PMFS (DRAM+NVM)

NUMA Interface – NVM- Only 12 Virtual CPU where all memory access uses the NVM portion of DRAM. No change to application code.

PMFS Interface – DRAM+NVM 13 Special filesystem designed for byte- addressable NVM. Avoids overhead of traditional filesystems.

DBMS Architectures 14 Disk-oriented. Main memory-oriented.

Disk-oriented DBMS 15 Pessimistic assumption that the data a txn needs is not in memory Based on the design assumptions made in the 1970s. –Ingres (Berkeley) –System R (IBM)

Primary Storage Buffer Pool Application ARIES LOG

Memory-oriented DBMS 17 Assume that all data fits in memory. Avoid the overhead of concurrency control + recovery. –SmallBase (AT&T) –Hekaton (Microsoft) –H-Store/VoltDB (Me & others…)

Snapshots Primary Storage Application CMD LOG

Experimental Evaluation 19 Compare the DBMS architectures on the two NVM interfaces. Yahoo! Cloud Serving Benchmark: –10 million records (~10GB) –8x database / memory –Variable skew

Evaluated Systems 20 NVM-Only –H-Store (v2014) –MySQL (v5.5) NVM+DRAM –H-Store + Anti-Caching (v2014) –MySQL (v5.5)

21 YCSB // TXN/SEC H-Store NUMA Interface (NVM-Only) Read-Only Workload 2x Latency Relative to DRAM SKEW AMOUNT (HIGH → LOW) MySQL

22 YCSB // TXN/SEC Anti-Caching PMFS Interface (NVM+DRAM) Read-Only Workload 2x Latency Relative to DRAM SKEW AMOUNT (HIGH → LOW) MySQL

23 YCSB // TXN/SEC NUMA Interface (NVM-Only) Write-Heavy Workload 2x Latency Relative to DRAM SKEW AMOUNT (HIGH → LOW) H-Store MySQL

24 YCSB // TXN/SEC PMFS Interface (NVM+DRAM) Write-Heavy Workload 2x Latency Relative to DRAM SKEW AMOUNT (HIGH → LOW) Anti-Caching MySQL

Discussion 25 NVM latency did not make a big difference in performance. Logging is major bottleneck in DBMS performance on NVM. MySQL wastes NVM space.

Database CMU

Database CMU 27 N-Store H-Store + Anti-Caching ThomasDB MongoDB

N-STORE

N-Store 29 NVM-only Architecture. Hybrid OLTP/OLAP DBMS: –High-performance txn processing. –Low-latency analytical operations. –Instant recovery.

30 N-Store – Shadow Paging Shadow Page Table Master Page Table DB Root X X X X

H-STORE

H-Store + Anti-Caching 32 Allows DBMS to mange DBs that are larger than amount of DRAM. Reducing memory overhead of evicted data. Using ML to reduce disk I/O.

MAIN TUPLE TABLE HEADER { EVICTED TUPLE TABLE INDEXES ANTI-CACHE Ø ANTI-CACHE BLOCKSLRU CHAIN 33

H-Store + Anti-Caching 34 Bloom filter usage tracking. Multi-faceted indexes. Cold storage approximations. Semantic block clustering.

ThomasD B

36 High-performance, low- overhead incremental computation platform. Maintains mapping between function invocations and data.

37 ThomasDB – Data Marts PUBLIC DATABASE PRIVATE DATABASE UDF Update! Materialized View ANALYTICS PROGRAM

38 ThomasDB – Preliminary Results MAP FUNCTION OVER WIKIPEDIA CORPUS

MongoDB

Distributed Document Database Design 40 Automatic tool selects the near-optimal physical design: –Sharding Keys –Denormalization –Indexes

Vigilante DBA 41 Automated framework for finding DB applications and fixing them. Build a large catalog of database applications from public sources.

Summary 42 Lots of database stuff in the works. Always looking for industry collaborators.

@ANDY _ PAVLO END db.cs.cmu.edu

This talk is in compliance with the Federal Bureau of Prisons (FBP) Act (1930) Pub. L. No , 46 Stat All reference herein to any specific commercial products, process, or service by trade name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply its endorsement, recommendation, or favoring by the FBP. The views and opinions of authors expressed herein do not necessarily state or reflect those of the FBP, and shall not be used for evaluation by including, but not limited, to FBP parole board, its directors, and other corrections officers. Regulations are contained in Title 28, Chapter V of the Code of Federal Regulations (CFR). Contact with parolees for research questions is regulated under 28 C.F.R All complaints about the validity of said research should be directed to the 28 C.F.R. 542 Administrative Remedy Program for a formal review of an issue relating to any aspect of the evaluation that "does not apply to inmates confined in other non-federal facilities."

Andy Pavlo Col. Stan Zdonik Mike Stonebraker Justin DeBrabant Joy Arulraj Subramanya Dulloor Rajesh Sankaran Jeff Parkhurst