c-treeACE V10.0 Update Guide

Performance

Scale UP to Enhanced Performance!

Commit Read Lock optimizations Advanced hash lookups speed filename searches Unbuffered I/O bypasses file system cache Faster numbering of transactions

Reduced memory suballocator contention scalability Reader/Writer Locks enhance synchronization efficiency Numerous SQL query optimizations

Range search optimizations Unix shared-memory connections Reduced index node contention Critical Section Spin Count optimizations on Windows

Reduced file header reads Intelligent file open/close logic for increased efficiency of file handlers Improved PreImage search of transaction hash entries

30% Faster Transaction Throughput	60% SQL Performance Improvement	200% Better Replication Throughput	26% Faster Read Performance Testing
Test Environment: FairCom’s cttctx load test utility simulating a record add / read / delete sequence on 23 files with indices.	Test Environment: Benchmark Factory TPC‑B	Test Environment: FairCom’s cttctx load test simulating a record add/read/delete on 23 files with indices then applying those changes to a target server on the same hardware.	Test Environment: FairCom’s cttctx load test utility simulating a record read sequence on 23 files with indices.
• 100 threads • 10,000 iterations • 1,000,000 transactions	• 20 threads • 3 minute run • 1.5GB File size	• 100 threads • 10,000 iterations • 1,000,000 transactions	• 100 threads • 10,000 iterations • 1,000,000 transactions
• Hardware: Dell PowerEdge, 16 Cores, 3.6 GHz, 32 GB RAM, Seagate ST3600057SS SCSI Hard Drive, 2 X 600GB, 15K rpm • Windows Server 2008 R2 64-bit, Datacenter Service Pack 1

The considerable performance gains experienced with the V10 release are a result of astute profiling of the c-treeACE database engine internals. Scalability issues, bottlenecks, contention points, wait times, capacity sizes, as well as many other details were wholly critiqued through thorough profiling, testing, and extensive benchmarking. Additionally we encouraged and considered “real world” feedback from customers. We are excited with the results and encourage you to adopt this V10 technology as soon as possible and experience these rewards.

Later in these notes, the Foundations of V10 Performance Gains section details the significant internal improvements within the c-treeACE core engine. Many are low-level internal changes and the combination of all these improvements resulted in overall higher c-treeACE performance with much improved scalability. The highlights are summarized below. The internal details are most applicable to those who wish to understand the depth behind the significant performance improvements in V10.

FairCom has subjected our latest releases to extensive testing in areas of scalability and performance to leverage technological advances in today’s operating systems and hardware platforms. We have achieved significant internal core engine performance improvements in the areas listed below.

Synchronization Objects - Semaphore/Mutex:

• Shared memory control
• Reduced synchronization redundancies
• Minimized internal contention of read/write queues
• Profiled memory file inserts
• Improved critical section spin count control
• Reduced node contention in index key lock management
• Efficiencies gained through native OS “reader/writer” synchronization support

Transaction Processing:

• Faster Automatic Recovery
• Faster Transaction Log processing
• Faster assignment of Transaction Numbers
• SAVE-POINT efficiency
• Transaction Preimage performance gains
• Dynamic Hash Bins: transaction entry search speed

File I/O:

• File Read-Only Open speed: Large number of open files
• File Open/Close overhead: Leave files open
• File name Hash List improvements
• Faster internal user file control block handling

Memory:

• Memory sub-allocator scalability improvements
• Memory sub-allocator adjustments for faster memory searches
• Memory File Configurable Hash Bins: Improved 64-bit addressing

Locking:

• Commit Read Locks: Significant performance gains; enhanced scalability
• Exclusive Files: No lock overhead
• Key-Level Lock optimizations

More Advances:

• User connection performance during a live backup: Dynamic Dump
• Estimate Key efficiencies
• Range search speed
• Cache efficiency: Unbuffered I/O to bypass OS's file system cache
• Read Header read: minimize need; increase speed

Need more proof? For a detailed list of the development work we have done to bring you enhanced performance, see the section titled Foundations of V10 Performance Gains.