Mohamed KEITA
Note #9•4 min read

Africa: The Continent Forgotten by Database Benchmarks

Performance benchmarks play a central role in the evaluation of storage engines and distributed systems. Tools like YCSB, TPC-C, sysbench, and others have shaped the industry’s understanding of “fast” vs “slow,” “scalable” vs “limited,” and “efficient” vs “inefficient.”
But all benchmarks embed assumptions about hardware, network stability, latency, access patterns, and workload structure.

And these assumptions simply do not reflect African realities.

This note explores why mainstream benchmarks fail to model the environments, constraints, and workloads found across African industries. It also highlights real-world examples (mobile money, logistics, e-government platforms, and distributed retail) that require a fundamentally different approach to system evaluation.

The Implicit Assumptions Behind Popular Benchmarks

Benchmark suites were designed with specific environments in mind — typically North America, Europe, and parts of Asia.
They implicitly assume:

* stable datacenter-grade networks
* abundant CPU and RAM
* reliable power supply
* SSD or NVMe storage
* minimal packet loss
* predictable workloads

Workloads in these regions tend to be:

  • high throughput,
  • predictable read/write ratios,
  • hosted in centralized datacenters,
  • driven by web or transactional traffic with stable connectivity.

Under these assumptions, benchmarks behave consistently and meaningfully.
But the moment these conditions change as they do across much of Africa benchmark conclusions no longer apply.

Why These Benchmarks Do Not Represent African Workloads

African digital ecosystems combine high demand, unpredictable connectivity, and distributed usage patterns.
This produces workloads fundamentally different from what TPC-C, YCSB, or sysbench were designed to model.

1. WAN Latency Is a First-Class Constraint

Benchmark workloads are almost always measured inside a datacenter.
But in Africa:

Local ISP variability → large jitter
Packet loss → 1–5%

Benchmarks assume sub-millisecond internal communication.
African deployments operate on global-scale latencies, often over unstable paths.

2. Workloads Are Geographically Distributed by Design

Benchmarks assume centralized infrastructure.
But African usage patterns involve:

  • multi-branch institutions,
  • rural points of service,
  • intermittently connected devices,
  • regional offices syncing over unreliable links.

These patterns are not represented in any standard benchmark.

3. Access Patterns Are Highly Skewed and Unpredictable

Typical benchmark distributions (uniform, Zipfian) cannot model:

  • spiky transaction bursts (e.g., salary day mobile money),
  • long periods of low activity punctuated by massive surges,
  • network partitions causing delayed write batches,
  • local caching followed by sync storms.

Benchmarks measure engines — African workloads measure resilience under irregularity.

4. Benchmarks Ignore Offline and Local-First Scenarios

In Africa, offline-tolerant systems are not luxuries — they are necessities.
Standard benchmarks do not account for:

* conflict reconciliation
* sync-at-once spikes
* intermittent batch replication
* temporary isolation of nodes

These factors fundamentally reshape the behavior of storage engines.

Real African Workloads That Benchmarks Fail to Model

Mobile Money

Mobile money systems process:

  • millions of low-value transactions,
  • high peak loads,
  • strict consistency constraints,
  • highly uneven geographic usage,
  • branches and kiosks with unstable connectivity.

A typical salary-day spike cannot be modeled with TPC-C.

Logistics & Transport

African logistics companies synchronize:

  • vehicle positions,
  • delivery events,
  • routing updates,
  • warehouse operations

across wide, unreliable geographies.
Benchmarks assume stable LAN environments — logistics requires survivability under fragmentation.

E-Government Platforms

Public administration systems must operate across:

  • remote offices,
  • rural municipalities,
  • variable connectivity,
  • heterogeneous hardware.

A centralized cloud-only model collapses under these constraints.

Distributed Retail (POS Systems)

Retail operations rely on:

  • local caching,
  • offline queueing of sales,
  • periodic synchronization,
  • conflict resolution.

None of these behaviors appear in YCSB or sysbench.

Conclusion

Benchmarks are not neutral.
They encode assumptions about infrastructure that reflect the environments in which they were created. Not those in which they are deployed.

Africa is structurally absent from these assumptions.
Its connectivity patterns, distribution models, peak events, and offline constraints demand new evaluation methodologies.

To build systems for Africa, we must benchmark like Africa.

Recommended References

  1. TPC-C Benchmark Specification
  2. YCSB — Core Workload Definitions
  3. ACM Queue — The Tail at Scale
  4. GSMA — Mobile Money Reports
  5. World Bank — Digital Infrastructure in Sub-Saharan Africa