Oracle vs Azure -- Performance Benchmarks¶

Query performance comparison: Oracle Database vs Azure SQL Managed Instance vs Azure Database for PostgreSQL. Transaction throughput, IOPS, concurrent session handling, and cost-per-transaction analysis.

These benchmarks are representative, not definitive. Actual performance depends on workload characteristics, data volume, indexing strategy, hardware configuration, and application patterns. Oracle performance on Exadata will differ significantly from Oracle on commodity hardware. Azure SQL MI performance varies by tier (General Purpose vs Business Critical) and vCore count. Always run your own benchmarks with your specific workload before making migration decisions.

1. Test environment¶

1.1 Configurations tested¶

1.2 Test database¶

• Size: 50 GB (100 tables, 500M rows total)
• Schema: OLTP workload (orders, customers, products, inventory)
• Indexes: Standard B-tree on primary keys, foreign keys, and common query columns
• Data distribution: Realistic skew (80/20 on popular products)

2. OLTP transaction throughput (TPC-C style)¶

2.1 New order transaction¶

The TPC-C new order transaction is a standard benchmark for OLTP databases. It involves reading from multiple tables, inserting into orders and order_lines, and updating inventory.

Key observations:

• Oracle Exadata delivers the highest raw throughput due to Exadata Smart Scan and Flash Cache
• Azure SQL MI Business Critical approaches Exadata performance at 20% of the cost
• Azure PostgreSQL provides the lowest cost-per-transaction
• For most federal OLTP workloads (< 5,000 TPS), all Azure targets provide adequate throughput

2.2 Mixed workload (70% read, 30% write)¶

3. Analytical query performance¶

3.1 Aggregation queries¶

Test query: SELECT department, SUM(amount), COUNT(*), AVG(amount) FROM transactions WHERE date >= '2024-01-01' GROUP BY department ORDER BY SUM(amount) DESC

Table size: 100M rows, 15 GB

3.2 Join-heavy analytical query¶

Test query: Multi-table join across orders, customers, products, inventory with 5 aggregations and filtering.

3.3 Window function performance¶

Test query: ROW_NUMBER, LAG, SUM OVER, PERCENTILE_CONT across 50M rows with multiple partitions.

For heavy analytical workloads, consider Fabric Mirroring + Direct Lake instead of running analytics on the OLTP database. Fabric's Spark engine processes analytical queries on OneLake data without impacting the transactional database. This is the CSA-in-a-Box pattern: OLTP database for transactions, Fabric for analytics.

4. IOPS and storage performance¶

4.1 Random read IOPS (8K blocks)¶

4.2 Sequential write throughput¶

5. Concurrent session handling¶

5.1 Connection scaling¶

Oracle DBAs are accustomed to dedicated server connections. Azure SQL MI and PostgreSQL benefit significantly from connection pooling. Use PgBouncer (built-in for Azure PostgreSQL) or application-level pooling to maintain performance at high concurrency.

6. Cost-per-transaction analysis¶

6.1 Cost efficiency at different workload levels¶

Cost includes: License (amortized for Oracle), compute, storage, HA, DR, backups, patching labor (for Oracle).

6.2 Break-even analysis¶

At what workload level does Oracle cost less per transaction than Azure?

Answer: Oracle never costs less per transaction because the fixed license + support cost dominates. Even at very high transaction volumes (1B+/month), Oracle's 22% annual support fee and DBA labor costs exceed Azure managed service pricing. The only scenario where Oracle has a cost advantage is when the licenses are already fully amortized (paid off) and only support is being paid -- and even then, support alone ($315K/year for a 32-core server) typically exceeds Azure SQL MI or PostgreSQL pricing for equivalent compute.

7. Scaling comparison¶

7.1 Vertical scaling¶

7.2 Horizontal scaling¶

8. Performance optimization tips for migrated workloads¶

8.1 Azure SQL MI¶

-- Enable automatic tuning
ALTER DATABASE FEDDB SET AUTOMATIC_TUNING
    (CREATE_INDEX = ON, DROP_INDEX = ON, FORCE_LAST_GOOD_PLAN = ON);

-- Add columnstore indexes for analytical queries
CREATE NONCLUSTERED COLUMNSTORE INDEX ncci_transactions
ON dbo.transactions (transaction_date, amount, department_id, product_id);

-- Review missing index recommendations
SELECT TOP 20
    mig.index_group_handle,
    mid.statement AS table_name,
    mid.equality_columns,
    mid.inequality_columns,
    mid.included_columns,
    migs.avg_user_impact
FROM sys.dm_db_missing_index_groups mig
JOIN sys.dm_db_missing_index_group_stats migs ON mig.index_group_handle = migs.group_handle
JOIN sys.dm_db_missing_index_details mid ON mig.index_handle = mid.index_handle
ORDER BY migs.avg_user_impact DESC;

8.2 Azure PostgreSQL¶

-- Tune key parameters for Oracle-migrated workloads
ALTER SYSTEM SET shared_buffers = '8GB';            -- 25% of RAM
ALTER SYSTEM SET effective_cache_size = '24GB';      -- 75% of RAM
ALTER SYSTEM SET work_mem = '64MB';                  -- For complex sorts/joins
ALTER SYSTEM SET maintenance_work_mem = '1GB';       -- For index builds
ALTER SYSTEM SET max_parallel_workers_per_gather = 4;-- Parallel query
ALTER SYSTEM SET jit = 'on';                         -- JIT compilation

-- Enable pg_stat_statements for query analysis
CREATE EXTENSION IF NOT EXISTS pg_stat_statements;

-- Find slow queries (equivalent of Oracle AWR top SQL)
SELECT query, calls, mean_exec_time::numeric(10,2) AS avg_ms,
       total_exec_time::numeric(10,2) AS total_ms,
       rows, shared_blks_hit, shared_blks_read
FROM pg_stat_statements
ORDER BY total_exec_time DESC
LIMIT 20;

9. Summary¶

For most federal OLTP workloads, Azure SQL MI Business Critical provides competitive performance at 20% of Oracle Exadata cost. For cost-optimized workloads, Azure PostgreSQL provides the best cost-per-transaction ratio. For analytics, Fabric + Direct Lake (via CSA-in-a-Box) outperforms any OLTP database for analytical queries by leveraging columnar storage and distributed compute.

Maintainers: csa-inabox core team Last updated: 2026-04-30