Weaving Relations for Cache Performance

Transcription

1 Weaving Relations for Cache Performance Anastassia Ailamaki Carnegie Mellon David DeWitt, Mark Hill, and Marios Skounakis University of Wisconsin-Madison

2 Memory Hierarchies PROCESSOR EXECUTION PIPELINE MAIN MEMORY

3 Memory Hierarchies PROCESSOR EXECUTION PIPELINE L1 I-CACHE L1 D-CACHE L2 CACHE $$$ MAIN MEMORY Cache misses are etremely epensive

4 Processor/Memory Speed Gap cycles per instruction VAX 11/780 CPI memory latency Pentium II Xeon Memory latency 1 access to memory # 1000 instruction opportunities

5 Breakdown of Memory Delays PII Xeon running NT 4.0, 4 commercial DBMSs: A,B,C,D Memory-related delays: 40%-80% of eecution time Range Selection (no inde) Range selection (clustered inde) Join (no inde) Memory stall time (%) 100% 80% 60% 40% 20% 0% 100% 80% 60% 40% 20% 0% 100% 80% 60% 40% 20% 0% A B C D DBMS A B C D DBMS A B C D DBMS L1 Data L2 Data L1 Instruction L2 Instruction Data accesses on caches: 19%-86% of memory stalls

6 Data Placement on Disk Pages Slotted Pages: Used by all commercial DBMSs 9 Store table records sequentially - Intra-record locality (attributes of record r together) / Doesn t work well on today s memory hierarchies Alternative: Vertical partitioning [Copeland 85] 9 Store n-attribute table as n single-attribute tables - Inter-record locality, saves unnecessary I/O / Destroys intra-record locality => epensive to reconstruct record Contribution: Partition Attributes Across - have the cake and eat it, too Inter-record locality + low record reconstruction cost

7 Outline The memory/processor speed gap What s wrong with slotted pages? Partition Attributes Across (PAX) Performance results Summary

8 Current Scheme: Slotted Pages Formal name: NSM (N-ary Storage Model) R PAGE HEADER RH ,' 661 1DPH -DQH $JH Jane 30 RH John 45 RH Jim 20 RH4 -RKQ 7658 Susan 52 -LP 6XVDQ /HRQ 'DQ Records are stored sequentially Offsets to start of each record at end of page

9 Predicate Evaluation using NSM PAGE HEADER RH Jane 30 RH John 45 RH Jim 20 RH4 Jane 30 RH block 1 45 RH block Susan Leon Jim 20 RH4 block Leon block 4 CACHE MAIN MEMORY VHOHFW QDPH IURP 5 ZKHUH DJH! NSM pushes non-referenced data to the cache

10 Need New Data Page Layout Eliminates unnecessary memory accesses Improves inter-record locality Keeps a record s fields together Does not affect I/O performance and, most importantly, is low-implementation-cost, high-impact

11 Partition Attributes Across (PAX) NSM PAGE PAX PAGE PAGE HEADER RH PAGE HEADER Jane 30 RH John RH Jim 20 RH Susan 52 Jane John Jim Susan Partition data within the page for spatial locality

12 Predicate Evaluation using PAX PAGE HEADER block 1 Jane John Jim Suzan CACHE VHOHFW QDPH IURP 5 MAIN MEMORY ZKHUH DJH! Fewer cache misses, low reconstruction cost

13 A Real NSM Record HEADER FIXED-LENGTH VALUES VARIABLE-LENGTH VALUES null bitmap, record length, etc offsets to variablelength fields NSM: All fields of record stored together + slots

14 PAX: Detailed Design # records free space attribute # attributes sizes pid v 4 f } Page Header F - Minipage Jane John presence bits v-offsets presence bits V - Minipage F - Minipage PAX: Group fields + amortizes record headers

15 Outline The memory/processor speed gap What s wrong with slotted pages? Partition Attributes Across (PAX) Performance results Summary

16 Sanity Check: Basic Evaluation Main-memory resident R, numeric fields Query: select avg (a i ) from R where a j >= Lo and a j <= Hi PII Xeon running Windows NT 4 16KB L1-I, 16KB L1-D, 512 KB L2, 512 MB RAM Used processor counters Implemented schemes on Shore Storage Manager Similar behavior to commercial Database Systems

17 Why Use Shore? Compare Shore query behavior with commercial DBMS Eecution time & memory delays (range selection) 100% Eecution time breakdown 100% Memory stall time breakdown % eecution time 80% 60% 40% 20% Memory stall time (%) 80% 60% 40% 20% 0% 0% A B C D Shore DBMS Computation Memory Branch mispr. Resource A B C D Shore DBMS L1 Data L2 Data L1 Instruction L2 Instruction We can use Shore to evaluate DSS workload behavior

18 Effect on Accessing Cache Data Cache data stalls Sensitivity to Selectivity stall cycles / record NSM PAX page layout L1 Data stalls L2 Data stalls stall cycles / record NSM L2 PAX L2 1% 5% 10% 20% 50% 100% selectivity PAX saves 70% of NSM s data cache penalty PAX reduces cache misses at both L1 and L2 Selectivity doesn t matter for PAX data stalls

19 Time and Sensitivity Analysis 1800 Eecution time breakdown 6 Sensitivity to # of attributes clock cycles per record Hardware Resource Branch Mispredict Memory Computation elapsed time (sec) NSM PAX NSM PAX page layout # of attributes in record PAX: 75% less memory penalty than NSM (10% of time) Eecution times converge as number of attrs increases

20 Evaluation Using a DSS Benchmark 100M, 200M, and 500M TPC-H DBs Queries: 1. Range Selections w/ variable parameters (RS) 2. TPC-H Q1 and Q6 sequential scans lots of aggregates (sum, avg, count) grouping/ordering of results 3. TPC-H Q12 and Q14 (Adaptive Hybrid) Hash Join comple where clause, conditional aggregates 128MB buffer pool

21 TPC-H Queries: Speedup PAX/NSM Speedup on PII/NT PAX/NSM Speedup 60% 45% 30% 15% 100 MB 200 MB 500 MB 0% RS Q1 Q6 Q12 Q14 Query PAX improves performance even with I/O Speedup differs across DB sizes

22 Updates Policy: Update in-place Variable-length: Shift when needed PAX only needs shift minipage data Update statement: update R set a p =a p + b where a q > Lo and a q < Hi

23 Updates: Speedup PAX/NSM Speedup on PII/NT PAX/NSM Speedup 18% 16% 14% 12% 10% 8% 6% 4% 2% 0% 2% 10% 20% 50% 100% Number of updated attributes PAX always speeds queries up (7-17%) Lower selectivity => reads dominate speedup High selectivity => speedup dominated by write-backs

24 PAX: a low-cost, high-impact DP technique Performance Eliminates unnecessary memory references High utilization of cache space/bandwidth Faster than NSM (does not affect I/O) Usability Summary Orthogonal to other storage decisions Easy to implement in large eisting DBMSs