Storage Format Lock-in: The Constraint Limiting Query Performance

Modern distributed processing engines achieve remarkable scale and reliability, yet they share a fundamental architectural constraint: rigid storage format requirements. This constraint forces organizations to accept significant performance penalties, sometimes 10-100x slower queries than technically possible, simply because their processing engine cannot adapt to optimized storage formats.

Why Parquet Falls Short for Selective Analytics

Parquet excels as a universal columnar format, but its design prioritizes compatibility and compression over query performance. Understanding its limitations requires examining how analytical queries actually execute.

The Row Group Problem

Parquet organizes data into row groups, typically 100,000 to 1 million rows each. This coarse granularity creates a fundamental problem: if even a single row in a row group matches your query filter, you must read ALL requested columns for the ENTIRE row group.

Consider a query filtering for a specific customer ID in a billion-row table. That customer's 100 transactions might be scattered across 100 different row groups. Parquet must read 100 row groups × 100,000 rows × all requested columns, processing 10 million rows to return 100.

Single-Phase Execution

Parquet readers process queries in a single phase:
‍

1. Read all requested columns for relevant row groups

2. Decompress the data

3. Apply filters

4. Return matching rows

This means reading and decompressing massive amounts of data that will immediately be discarded. There's no mechanism to filter first, then read only what's needed.

Limited Predicate Pushdown

While Parquet stores min/max statistics per row group and optional Bloom filters, these only help skip entire row groups. For analytical queries where some matching rows exist in many row groups, this provides minimal benefit. You still read entire 100,000-row chunks to extract perhaps 10 rows from each.

How Query-Optimized Formats Solve These Problems

Formats like ClickHouse's MergeTree take a fundamentally different approach:

Granular Storage

Instead of 100,000-row groups, data is organized into 8,192-row granules. This 12x finer granularity means reading much less unnecessary data when matches are sparse. Finding those same 100 customer transactions requires reading ~12x less data just from granularity alone.

Two-Phase Execution with PREWHERE

Query-optimized formats implement two-phase execution:

1. Phase 1: Read ONLY filter columns for candidate granules

2. Phase 2: For rows that pass filters, read the remaining requested columns

This seemingly simple change has a profound impact. Instead of reading 20 columns for millions of rows, you read 1-2 filter columns first, identify the 100 matching rows, then read the other 18 columns for just those 100 rows.

Sparse Indexing

A sparse primary index stores one entry per granule (every 8,192 rows), creating a tiny index that fits entirely in memory even for billion-row tables. Binary search on this index instantly identifies which granules to read, eliminating 99%+ of data before any I/O occurs.

Lazy Materialization

Column reads are deferred until absolutely necessary. If a query has multiple filters, the engine applies them progressively, reading additional columns only for rows that survive each filter. This minimizes decompression work and memory bandwidth usage.

The Performance Gap: A First-Principles Calculation

Let's calculate the actual performance difference using a realistic analytical query on data stored in S3:

Scenario:

• 100 million rows, 100 columns
• Query with 0.01% selectivity returning 10,000 rows
• Projects 20 columns
• All data in S3 (network I/O is the same for both formats)

Parquet Execution:

• Data organized in 100,000-row groups
• 10,000 matching rows spread across ~100 row groups
• Must read all 20 requested columns for entire row groups containing any match
• Compressed data transferred from S3: 160 MB (with 10:1 compression)
• Data decompressed and processed: 1.6 GB

Query-Optimized Format (like ClickHouse MergeTree):

• Data organized in 8,192-row granules with sparse index
• Two-phase execution: read filter columns first, identify matches, then read other columns
• Compressed data transferred from S3: 46 MB (with 3.5:1 compression)
• Data decompressed and processed: 162 MB

Despite storing data 2.9x larger on S3 due to less aggressive compression, the query-optimized format transfers 3.5x less data over the network and processes 10x less decompressed data. Processing 1.6 GB vs 162 MB of decompressed data is the difference between 100ms and 10ms query time

Why Processing Engines Don't Support Alternative Formats

While Spark and Ray technically could support arbitrary storage formats through extension APIs, in practice they don't. The ecosystem has converged on Parquet, ORC, and Avro, leaving significant performance opportunities unexplored.

The Spark Reality

Spark provides a DataSource V2 API for custom formats, yet virtually no production deployments use them. The practical barriers:

• Format implementations must handle Spark's complex internal row representation
• Custom formats cannot leverage Spark's vectorized execution optimizations
• The Catalyst optimizer cannot reason about custom format capabilities
• Maintaining custom format readers requires deep Spark internals knowledge

Ray's Format Limitations

Ray delegates data loading to Pandas or PyArrow:

python

@ray.remote

def process_partition(file_path):

    # Limited to formats Pandas supports

    df = pd.read_parquet(file_path)

    return df.groupby('device_id').mean()

Adding optimized formats would require writing C++ extensions, ensuring serialization compatibility, and maintaining format-specific optimizations Ray doesn't understand.

The Architectural Impedance Mismatch

Even when custom format support exists, fundamental architectural assumptions prevent leveraging format-specific optimizations. Standard engines read all requested columns before filtering and lack the execution primitives for two-phase execution. The granularity mismatch between coarse row groups and fine granules cannot be bridged through format plugins.

Real-World Impact on S3-Based Data Lakes

Production systems using S3 data lakes demonstrate concrete impact:

E-commerce Recommendation Engine

• Dataset in S3: 500M products × 800 attributes
• Parquet in Spark: 8.2 seconds average latency
• Possible with optimized format: 100ms
• Impact: Real-time recommendations impossible

Financial Risk Analytics

• Dataset in S3: 10 years of trades, 2,000 columns
• Parquet in Spark: 45 seconds per portfolio
• Possible with optimized format: 500ms
• Impact: Risk managers wait minutes for updates that could take seconds

IoT Monitoring

• Dataset in S3: 100,000 devices × 1,000 metrics
• Parquet in Ray: 12 seconds per device group
• Possible with optimized format: 50ms
• Impact: Alert latency prevents real-time response

The Compression-Performance Trade-off

Parquet achieves superior compression, typically 2-3x better than query-optimized formats. A 100GB dataset might compress to 10GB in Parquet versus 30GB in an optimized format, translating to $2.30/month versus $6.90/month on S3.

However, compute costs dominate this equation. When queries run 10-100x faster with optimized formats, the required infrastructure shrinks proportionally. A workload requiring a 16-node Spark cluster at $8,000/month might run on 2 nodes at $1,000/month with format optimization. The $7,000/month compute savings overwhelms the $4.60/month additional S3 storage cost by a factor of 1,500x.

Performance Patterns by Query Type

Different query patterns show varying sensitivity to format selection:

Highly Selective Queries (<0.1% of rows returned)

• Performance difference: 20-100x
• Parquet's coarse row groups create massive read amplification

Wide Table Projections (many columns, few rows)

• Performance difference: 10-50x
• Single-phase execution forces reading all columns upfront

Aggregation Queries

• Performance difference: 5-20x
• Processing unnecessary rows dominates computation

Full Table Scans

• Performance difference: 0.7-1.4x (Parquet often faster)
• Parquet's superior compression provides advantage when reading everything

The Hidden Costs of Format Lock-in

Beyond direct performance impact, format rigidity creates cascading inefficiencies:

Infrastructure Over-provisioning: Organizations scale clusters to compensate for format inefficiency. A workload naturally requiring 2 nodes might run on 20 nodes to achieve acceptable latency.

Architectural Workarounds: Teams implement pre-aggregation pipelines, materialized views, and caching layers, each adding operational complexity without addressing the root cause.

Opportunity Costs: When queries take 30 seconds instead of 300ms, entire categories of applications become infeasible.

The Path Forward

Parquet remains excellent for data interchange, archival storage, and full-scan workloads. The opportunity lies in enabling processing engines to leverage format diversity based on workload requirements.

The ideal architecture would support understanding the profile of the workloads, adjusting compaction accordingly, and using the best data format for the job. Current processing engines cannot implement this strategy due to their format rigidity. They lack the flexibility to transparently read different formats while applying format-specific optimizations.

This limitation is beginning to change. Next-generation execution engines recognize that format flexibility is essential for modern analytical workloads. By decoupling query execution from storage assumptions, systems like Flarion can deliver the performance that has always been technically possible but practically unreachable.

Organizations no longer need to accept 10x performance penalties as the price of distributed processing. The ability to match storage formats to query patterns represents the difference between queries that frustrate users and analytics that drive real-time decisions.

‍

Apache Spark 4.0 marks a significant milestone in the framework's evolution toward columnar processing. With enhanced Apache Arrow integration, improved UDF support, and refined plugin architectures, Spark has taken meaningful steps forward. Yet understanding both the advances and the remaining gaps reveals why the journey toward truly efficient columnar processing continues.

The Promise and Reality of Columnar Improvements

Spark 4.0's headline improvements center on Apache Arrow integration. The framework now offers direct DataFrame-to-Arrow conversions, Arrow-optimized Python UDFs achieving up to 1.9x performance improvements, and cleaner APIs for columnar data exchange. These changes particularly benefit PySpark users, who've long suffered from serialization overhead when moving data between JVM and Python processes.

Yet examining Spark's architecture reveals a fundamental reality: the core execution engine remains predominantly row-oriented. While Spark provides hooks for columnar execution through its plugin architecture, the built-in operators - projections, filters, joins, aggregations - still process data row by row through optimized Java code generation. This isn't an oversight but a deliberate design choice that prioritizes compatibility and stability over pure performance.

Where Columnar Support Actually Stands

Understanding Spark 4.0's columnar capabilities requires distinguishing between data format and execution model. Spark has long supported columnar storage formats like Parquet. However, during actual computation, most operations convert this columnar data back to rows for processing.

Built-in SQL expressions execute through Catalyst's code generation, producing tight Java loops that process one row at a time. Complex expressions involving conditionals, nested functions, or custom logic follow this row-wise pattern. The JVM's JIT compiler optimizes these loops well, but they fundamentally lack the vectorized operations that define true columnar processing.

UDF support presents a mixed picture. Pandas UDFs genuinely operate on columnar data, leveraging NumPy's vectorized operations. The new Arrow-optimized Python UDFs improve data transfer efficiency but don't change the scalar execution model - they still process individual values, just with better serialization. Scala and Java UDFs remain entirely row-based, forcing any columnar data to convert back to rows for execution.

The architectural split becomes clear when examining memory management. Spark uses its own ColumnVector implementations for internal operations, not pure Arrow format. Converting between Spark's internal format and Arrow involves either copying or wrapping data, adding overhead that pure columnar engines avoid.

The Performance Gap That Remains

The practical implications become evident in production workloads. Join operations still rely on sort-merge or hash algorithms implemented in Java without SIMD optimization. Aggregations process groups row by row rather than operating on entire column chunks. String operations, mathematical expressions, and date manipulations all follow the same pattern - optimized Java code that processes individual values rather than vectors.

Native columnar engines demonstrate what's possible with true vectorized execution. By leveraging SIMD instructions and processing entire column batches simultaneously, these engines achieve significant speedups - often 2x or more - on the same hardware. This isn't because Spark's code is poorly written; it's because columnar execution with hardware vectorization fundamentally outperforms row-wise processing for analytical workloads.

The memory efficiency gap proves equally significant. Native columnar engines process data in its compressed form, maintaining compression through operations wherever possible. Spark's row-wise operations require decompression and materialization, increasing memory pressure and triggering more frequent garbage collection. For workloads pushing memory limits - a common scenario given how frequently Spark jobs encounter OOM errors - this efficiency difference can determine whether jobs complete successfully.

The Path Forward: Complementary Solutions

Spark 4.0's columnar improvements represent genuine progress, particularly for Python workflows and data interchange scenarios. Yet the core execution engine's row-based nature means achieving optimal columnar performance requires additional components.

Organizations increasingly deploy hybrid architectures that leverage Spark's strengths - distributed orchestration, fault tolerance, broad connector support - while delegating performance-critical operations to specialized columnar engines. Whether through native code execution or hardware acceleration, these complementary technologies fill the gaps in Spark's columnar story. This is precisely where solutions like Flarion's Arrow-based columnar processing provide value - plugging directly into existing Spark deployments to accelerate workloads without requiring code changes, while maintaining the distributed capabilities teams already rely on.

Understanding both Spark 4.0's advances and its limitations enables informed architectural decisions. While Spark takes important steps toward columnar processing, the journey toward truly efficient columnar execution often requires recognizing where additional acceleration provides essential value. For teams facing today's performance challenges - growing datasets, tightening SLAs, and mounting infrastructure costs - combining Spark's orchestration capabilities with purpose-built columnar acceleration delivers the performance modern data platforms demand.

‍