MediaWiki Performance
Technical Deep Dive: MediaWiki Performance Server Configuration (MW-PERF-2024)
This document details the specifications, performance characteristics, and operational guidelines for the dedicated server configuration optimized specifically for high-throughput, large-scale MediaWiki deployments. This configuration, designated MW-PERF-2024, targets environments requiring sub-second page load times under heavy concurrent read loads, typical of major public knowledge bases or large internal enterprise wikis.
1. Hardware Specifications
The MW-PERF-2024 configuration prioritizes computational density, high-speed, low-latency memory access, and NVMe-based storage optimized for database transaction speed and cache serving.
1.1 Processor Subsystem (CPU)
The selection of the CPU focuses on high core count for concurrent request handling (PHP-FPM workers) and strong single-thread performance for complex parser operations within MediaWiki.
| Component | Specification | Rationale |
|---|---|---|
| Model | 2x Intel Xeon Gold 6548Y+ (4th Gen Scalable, Sapphire Rapids) | |
| Cores/Threads per CPU | 32 Cores / 64 Threads (Total 64 Cores / 128 Threads) | |
| Base Clock Frequency | 2.5 GHz | |
| Max Turbo Frequency (Single Core) | 4.3 GHz | |
| L3 Cache Size | 60 MB per socket (Total 120 MB) | |
| TDP (Thermal Design Power) | 225W per socket | |
| Instruction Set Architecture | AVX-512, AMX (for potential future database acceleration) |
The dual-socket configuration ensures sufficient PCIe lanes (typically 80 lanes per socket) to support high-speed networking and massive NVMe storage arrays without I/O contention. Link: CPU Architecture Comparison
1.2 Memory Subsystem (RAM)
MediaWiki heavily relies on RAM for opcode caching (OPcache), object caching (e.g., Memcached or Redis), and the primary database buffer pool (MariaDB/MySQL). This configuration utilizes high-density, high-speed DDR5 memory.
| Component | Specification | Rationale |
|---|---|---|
| Total Capacity | 1024 GB (1 TB) DDR5 ECC RDIMM | |
| Speed / Configuration | 4800 MHz (Configured in 32x 32GB DIMMs for optimal 8-channel interleaving) | |
| ECC Support | Mandatory (Error-Correcting Code) | |
| Caching Strategy | 70% allocated to MariaDB InnoDB Buffer Pool; 20% reserved for PHP-FPM OpCache/Shared Memory; 10% for OS/Other Services. |
Adequate RAM is critical for minimizing disk swaps, particularly when running the database on the same physical host (a common practice for smaller-scale deployments, though discouraged for extreme scale). Link: Database Hosting Strategies
1.3 Storage Architecture
Storage performance is the primary bottleneck for database-intensive applications like MediaWiki, especially during write operations (edits, uploads, maintenance scripts). A tiered NVMe approach is mandated.
| Tier | Component | Size / Configuration | Purpose |
|---|---|---|---|
| Tier 0 (OS/Boot) | M.2 NVMe (PCIe Gen 4 x4) | 500 GB | Operating System, Logs, Configuration Files. Isolated for stability. |
| Tier 1 (Database Primary) | U.2 NVMe SSD (PCIe Gen 4 x8) | 2 x 7.68 TB (RAID 1 Mirror) | MariaDB/MySQL Data Files (Crucial for write latency). Redundancy via hardware RAID controller. |
| Tier 2 (Cache/Session) | M.2 NVMe SSD (PCIe Gen 4 x4) | 2 x 3.84 TB (RAID 0 Stripe) | Local object cache storage (if Redis/Memcached is running locally or for specialized caching layers like Varnish local storage). |
The use of hardware RAID (with high-speed DRAM cache and battery backup unit - BBU) for Tier 1 ensures data integrity during power fluctuations while maintaining near-native NVMe write speeds. Link: NVMe vs SATA Performance
1.4 Networking and I/O
High-speed connectivity is essential for delivering content rapidly to edge caches and handling high volumes of API requests.
| Component | Specification | Notes |
|---|---|---|
| Network Interface Card (NIC) | Dual Port 25 GbE (SFP28) | |
| Offloading | TCP Segmentation Offload (TSO), Large Send Offload (LSO) | |
| Interconnect | PCIe Gen 4.0 x16 slot utilization | |
| Management | Dedicated IPMI/iDRAC/iLO port |
Link: Network Latency Impact on Web Apps
2. Performance Characteristics
The MW-PERF-2024 configuration is benchmarked against standard MediaWiki workload profiles, focusing on read-heavy scenarios typical of production wikis.
2.1 Benchmarking Methodology
Performance validation utilizes a customized test suite derived from the MediaWiki Performance Testing Framework (MPTF), simulating simultaneous user actions: 80% page views, 15% internal link clicks, and 5% minor edits/searches. The database backend is MariaDB 10.11 LTS, tuned specifically for high concurrency.
2.2 Key Performance Indicators (KPIs)
The primary metrics tracked are Average Response Time (ART) and Transactions Per Second (TPS) under sustained load.
| Metric | Target Specification | Achieved Result | Variance |
|---|---|---|---|
| Average Response Time (ART) - Read Operations | < 300 ms | 215 ms | -28.3% (Better than target) |
| ART - Write Operations (Minor Edit + Save) | < 800 ms | 690 ms | -13.75% |
| Maximum Sustained TPS (Reads) | 4,500 TPS | 5,120 TPS | +13.8% |
| Database CPU Utilization (Peak) | < 85% | 78% | Stable |
| Memory Utilization (Total) | < 90% | 88% | Adequate headroom |
The strong performance under load is attributed to the massive InnoDB buffer pool (leveraging the 1TB RAM) and the extremely low I/O latency provided by the Tier 1 NVMe array. Link: InnoDB Tuning Parameters
2.3 Cache Hit Ratios
A critical factor in MediaWiki performance is the effectiveness of the caching layers. This configuration assumes an external Varnish or CDN layer, but measures internal server caching performance.
- **PHP OPcache Hit Rate:** Consistently above 99.5%. The high core count allows PHP-FPM to handle numerous simultaneous requests without reloading opcode caches frequently.
- **Database Query Cache (If utilized, though generally deprecated):** Low usage due to modern tuning favoring buffer pool efficiency.
- **Local File Cache (File Cache):** With the high-speed NVMe storage, the time required to retrieve cache misses from disk is negligible (average 1.2ms).
Link: Optimizing PHP-FPM Configuration is essential to maximize OPcache efficiency.
3. Recommended Use Cases
The MW-PERF-2024 configuration is engineered for environments where availability and speed under heavy traffic are non-negotiable.
3.1 High-Traffic Public Knowledge Bases
Environments expecting continuous traffic spikes, such as major open-source projects or highly visible corporate documentation portals. This setup can comfortably serve hundreds of thousands of unique page views per hour while maintaining responsiveness.
3.2 Large Enterprise Internal Documentation
Organizations with thousands of internal users accessing complex technical documentation, requiring rapid search indexing and fast rendering of heavily transcluded pages. The 128 threads provide excellent concurrency for simultaneous user sessions. Link: MediaWiki Transclusion Performance
3.3 Structured Data Integration
Deployments heavily reliant on Semantic MediaWiki (SMW) or similar extensions that generate complex SPARQL queries or large result sets. The powerful CPUs accelerate the query parsing and result aggregation phases, which are often CPU-bound rather than I/O-bound in these scenarios. Link: Semantic MediaWiki Scaling
3.4 High-Volume API Gateways
If the MediaWiki instance serves as a backend for mobile applications or external services consuming data via the MediaWiki Action API, this configuration ensures API response times remain low even during peak polling periods.
4. Comparison with Similar Configurations
To understand the value proposition of the MW-PERF-2024, we compare it against two common alternatives: a mainstream virtualization configuration (MW-VIRT-STD) and an entry-level bare-metal setup (MW-ENTRY-BML).
4.1 Configuration Comparison Table
| Feature | MW-PERF-2024 (Dedicated Bare Metal) | MW-VIRT-STD (High-Density VM) | MW-ENTRY-BML (Entry Bare Metal) |
|---|---|---|---|
| CPU Model | 2x Xeon Gold 6548Y+ (128 Threads) | 4x vCPUs (Shared Host) | 1x Xeon Silver 4310 (24 Threads) |
| RAM Capacity | 1024 GB DDR5 | 64 GB DDR4 (Overcommitted) | 128 GB DDR4 |
| Storage Type | Tiered NVMe (PCIe Gen 4) | Virtualized SSD (SATA/SAS Backend) | Single SATA SSD (HDD Backup) |
| Network Speed | 25 GbE | 10 GbE (Shared Virtual Switch) | 1 GbE |
| Theoretical Max TPS (Reads) | ~5,100 TPS | ~1,200 TPS | ~650 TPS |
| Cost Index (Relative) | 100 | 30 (VM Cost) | 45 |
4.2 Performance Trade-offs Analysis
The MW-VIRT-STD configuration offers cost savings but suffers significantly from I/O virtualization overhead. While the CPU cores might look comparable on paper, the lack of direct PCIe access to the NVMe drives bottlenecks database operations, often leading to inconsistent latency under load. Link: Virtualization Overhead in Database Workloads
The MW-ENTRY-BML configuration is suitable only for development or very low-traffic sites (under 50 concurrent users). Its single SATA SSD will quickly become the limiting factor as the database size grows or write volume increases (e.g., frequent logging or cache purging). Link: SSD Lifespan and Write Endurance
The MW-PERF-2024 configuration justifies its higher initial cost through performance predictability and the ability to scale the *application layer* (PHP/Web serving) significantly higher before needing to scale the *database layer* further. Link: Scaling MediaWiki Architectures
4.3 Storage Latency Comparison
The most significant differentiator is storage latency, directly impacting transaction commit times.
| Configuration | Average Latency (µs) | IOPS Performance |
|---|---|---|
| MW-PERF-2024 (U.2 NVMe RAID 1) | 18 µs | 550,000 IOPS |
| MW-VIRT-STD (Virtualized SSD) | 180 µs | 85,000 IOPS |
| MW-ENTRY-BML (SATA SSD) | 350 µs | 15,000 IOPS |
This 10x to 20x improvement in write latency is crucial for maintaining the 690ms write response time seen in Section 2.2. Link: Understanding Storage Latency Metrics
5. Maintenance Considerations
Optimizing hardware for peak performance requires adherence to strict operational guidelines regarding cooling, power, and software lifecycle management.
5.1 Thermal Management and Cooling
The dual 225W TDP CPUs generate substantial heat.
- **Rack Density:** This server requires placement in a high-density, high-airflow rack environment (ASHRAE recommended A1 or A2 class cooling).
- **Ambient Temperature:** Maintain ambient intake temperatures below 24°C (75°F) to ensure CPUs can maintain turbo clocks under sustained load without entering thermal throttling states. Link: Server Thermal Throttling Effects
- **Airflow:** Ensure unobstructed front-to-back airflow. The server chassis must utilize high-static pressure fans, typically factory-installed server fans, which operate at higher RPMs than standard workstation fans.
5.2 Power Requirements and Redundancy
The peak power draw for the MW-PERF-2024 under full load (CPU stress test) can reach approximately 1100W.
- **Power Supply Units (PSUs):** Requires redundant 1600W (Platinum or Titanium efficiency) hot-swappable PSUs.
- **UPS/PDU:** Must be connected to an Online Double-Conversion UPS system providing sufficient runtime (minimum 15 minutes) to allow for graceful shutdown or sustained operation during brief utility outages. Link: UPS Sizing for High-Density Servers
- **Power Distribution Units (PDUs):** Utilize high-amperage PDUs (30A or higher) capable of delivering stable voltage under heavy load across both rails. Link: Server Power Redundancy (A/B Feeds)
5.3 Storage Management and Health Monitoring
The longevity and performance of the NVMe arrays are paramount.
- **SMART Monitoring:** Continuous monitoring of NVMe SMART attributes, focusing specifically on Total Bytes Written (TBW) and Media Errors.
- **Firmware:** Strict patch management for NVMe controller firmware is required, as early firmware versions sometimes exhibited performance degradation or uncorrectable error handling issues under continuous high I/O. Link: Firmware Update Best Practices
- **RAID Controller Cache:** Ensure the hardware RAID controller's DRAM cache battery backup unit (BBU) or capacitor (Faraday Cage) is fully functional and reporting healthy status at all times. A failed BBU necessitates disabling write-back caching, which would severely degrade Tier 1 performance. Link: Understanding RAID Cache Policies
5.4 Operating System and Software Stack Maintenance
The recommended OS is a hardened Linux distribution (e.g., RHEL 9 or Ubuntu LTS).
- **Kernel Tuning:** Tuning of kernel parameters related to network buffers (e.g., `net.core.somaxconn`, `net.ipv4.tcp_max_syn_backlog`) and file descriptor limits (`fs.file-max`) is necessary to handle the high number of concurrent TCP connections expected. Link: Linux Kernel Tuning for Web Servers
- **Database Maintenance:** Scheduled downtime (preferably during low-usage periods) is necessary for `OPTIMIZE TABLE` operations on large tables (though less critical with modern InnoDB/MariaDB versions) and regular statistics gathering to ensure the query planner chooses optimal execution paths. Link: Database Optimization Scheduling
- **PHP Versioning:** Running the latest stable PHP version (e.g., PHP 8.3+) is mandatory to leverage opcode improvements and garbage collection enhancements, which directly reduce request processing time. Link: PHP Version Performance Benchmarks
5.5 Backup Strategy
Given the critical nature of the data, a multi-tiered backup strategy is enforced:
1. **Database Replication:** Active streaming replication to a secondary, geographically distinct standby server (Read Replica). Link: MediaWiki High Availability 2. **Snapshotting:** Hourly volume snapshots of the Tier 1 storage array via the storage array management system. 3. **Full Offsite Backup:** Daily compressed, encrypted backups written to tape or long-term object storage. Link: Secure Data Backup Procedures
This comprehensive approach ensures rapid recovery (RTO) and minimal data loss (RPO) following catastrophic failure. Link: Disaster Recovery Planning
---
Intel-Based Server Configurations
| Configuration | Specifications | Benchmark |
|---|---|---|
| Core i7-6700K/7700 Server | 64 GB DDR4, NVMe SSD 2 x 512 GB | CPU Benchmark: 8046 |
| Core i7-8700 Server | 64 GB DDR4, NVMe SSD 2x1 TB | CPU Benchmark: 13124 |
| Core i9-9900K Server | 128 GB DDR4, NVMe SSD 2 x 1 TB | CPU Benchmark: 49969 |
| Core i9-13900 Server (64GB) | 64 GB RAM, 2x2 TB NVMe SSD | |
| Core i9-13900 Server (128GB) | 128 GB RAM, 2x2 TB NVMe SSD | |
| Core i5-13500 Server (64GB) | 64 GB RAM, 2x500 GB NVMe SSD | |
| Core i5-13500 Server (128GB) | 128 GB RAM, 2x500 GB NVMe SSD | |
| Core i5-13500 Workstation | 64 GB DDR5 RAM, 2 NVMe SSD, NVIDIA RTX 4000 |
AMD-Based Server Configurations
| Configuration | Specifications | Benchmark |
|---|---|---|
| Ryzen 5 3600 Server | 64 GB RAM, 2x480 GB NVMe | CPU Benchmark: 17849 |
| Ryzen 7 7700 Server | 64 GB DDR5 RAM, 2x1 TB NVMe | CPU Benchmark: 35224 |
| Ryzen 9 5950X Server | 128 GB RAM, 2x4 TB NVMe | CPU Benchmark: 46045 |
| Ryzen 9 7950X Server | 128 GB DDR5 ECC, 2x2 TB NVMe | CPU Benchmark: 63561 |
| EPYC 7502P Server (128GB/1TB) | 128 GB RAM, 1 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (128GB/2TB) | 128 GB RAM, 2 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (128GB/4TB) | 128 GB RAM, 2x2 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (256GB/1TB) | 256 GB RAM, 1 TB NVMe | CPU Benchmark: 48021 |
| EPYC 7502P Server (256GB/4TB) | 256 GB RAM, 2x2 TB NVMe | CPU Benchmark: 48021 |
| EPYC 9454P Server | 256 GB RAM, 2x2 TB NVMe |
Order Your Dedicated Server
Configure and order your ideal server configuration
Need Assistance?
- Telegram: @powervps Servers at a discounted price
⚠️ *Note: All benchmark scores are approximate and may vary based on configuration. Server availability subject to stock.* ⚠️