In this presentation Dmitry Yemanov, lead architect of Firebird Project, introduces the key advantages of upcoming major release of Firebird 3.

1. Firebird 3.0
Initial Review

2. Development priorities
● Primary goals
 SMP and multi-core friendly multi-threaded engine
with the shared cache
 Renewed architecture for the next versions
● Secondary goals
 Improved performance and scalability
 More security options
 Better query optimization
 Extended monitoring abilities
 SQL language enhancements

3. Shared cache
● Must have
 Shared page cache with scalable synchronization
 Support for both internal (cross-thread) and external
(cross-process) locking
● Nice to have
 Plan A: shared metadata cache,
shared compiled requests cache
 Plan B: no shared metadata,
but reduced memory usage
(release cached requests as soon as desirable)

4. Cache Architecture
Process 1 Process 2 Process N
Thread 1 Thread 1 Thread 1
Locking Locking
Thread 2 Thread 2 Thread 2
Thread N Thread N Thread N
Shared Cache Shared Cache Shared Cache
Database

5. Server Architecture
● Unified but flexible
 Engine library supporting single major ODS plus
server (aka network listener) executable
loading the engine library
 Shared vs exclusive database access:
configurable (per database) via firebird.conf
 Execution models available for the shared access
and configurable at the server level (using a
command-line switch):
– process based (former Classic)
– thread based (former SuperClassic)
 Embedded access may also be shared or exclusive

6. Server Architecture
Network
Server
(network listener)
Dispatcher
(Y-valve)
Legacy engine V3.0 engine Remote redirector
provider provider provider
Network
ODS11 ODS12
database database

7. Architecture in Older Terms
● SuperServer
 Exclusive database access, multi-threaded mode
 Maximum performance
● SuperClassic
 Shared database access, multi-threaded mode
 Allow other applications to access the database
● Classic
 Shared database access, multi-process mode
 Maximum reliability (crash tolerance)

8. ODS 12
● Physical structure
 Flag to mark data pages as not having any garbage
(improves sweep performance)
 Inventory of the page ↔ SCN relationships
(improves physical backup performance)
 RLE compression runs are extended from 128
bytes to 32K (less storage for declared long but de-
facto short or empty strings)
● Logical structure
 New system and monitoring tables

9. Optimization
● Algorithms
 Cost based approach wherever possible
 Efficient joining for streams containing aggregates,
unions and procedures
 INNER JOIN for dependent (via input parameters)
procedures now works properly
 Hash joins, outer merge joins, indexed full outer
joins
● Cached invariant subqueries
 Advanced (much more detailed) plan output

10. Advanced plan output
SELECT statement
-> First [10]
-> Sort [SUM, O_ORDERDATE]
-> Aggregate
-> Sort [L_ORDERKEY, O_ORDERDATE, O_SHIPPRIORITY]
-> Inner Loop Join
-> Filter
-> Table «ORDERS» Access By ID
-> Bitmap
-> Index «ORDERS_ORDERDATE» Range Scan
-> Filter
-> Table «CUSTOMER» Access By ID
-> Bitmap
-> Index «CUSTOMER_PK» Unique Scan
-> Filter
-> Table «LINEITEM» Access By ID
-> Bitmap
-> Index «LINEITEM_PK» Unique Scan

11. Optimization
● Statistics
 Table level data: page count, row count, average
compression ratio, etc
 Index level data: tree depth, page count, average
compression ratio, clustering factor, etc
 Value distribution histograms
 Refresh detalization: complete (column level) or
partial (table level)
 Manual or automatic (background) refreshing

12. SQL features
● Already implemented
 External procedures, functions and triggers
 PSQL scalar functions
 PSQL packages (similar to Oracle)
 DDL triggers
 Identity columns
 Window functions
 Scrollable (bi-directional) cursors in PSQL
 Parameterized user-defined exceptions

13. SQL features
● Planned
 Bi-directional cursors in API
 Enhanced (fixed) long numeric arithmetics
in Dialect 3
 Basic optimizer hints
 The «unstable cursor» issue to be fixed
 Timeouts for queries, transactions, connections
 Extended length limit for SQL queries, plans and
result sets

14. Security
● Network level
 Pluggable authentication
 Over-the-wire encryption
● Database level
 Optional per database authentication
(users stored inside the database)
 Page level encryption
 GRANT ROLE TO ROLE
 Implicit roles (aka user groups)
 Permissions for DDL operations

15. Release Schedule
● Maintenance
 2.0.7 — Q2-Q3 2011,
likely to be the last one in the v2.0.x series
 2.1.4 — Q1 2011,
2.1.5 to appear the next year
● New versions
 2.5.0 has been released,
2.5.1 to follow Q2 2011
 3.0 Alpha 1 — Q3 2011

16. Questions?