Aceso: Carbon-Aware and Cost-Effective Microservice Placement for Small and Medium-sized Enterprises

Aceso is an adaptive placement system designed for small and medium-sized enterprises that dynamically schedules microservices across geographically constrained regions to significantly reduce carbon emissions and operational costs while meeting latency requirements, specifically addressing the limitations of existing solutions that assume access to global-scale infrastructure.

The Gist

Imagine you run a busy, modern restaurant (your microservice application). Instead of one giant kitchen, you have many small, specialized stations: a grill for burgers, a salad bar, a drink station, and a dessert counter. Each station is a "microservice." They talk to each other constantly to get your order ready.

Now, imagine you want to be a "green" restaurant. You want to cook your food using electricity from the cleanest, most eco-friendly power plants possible. But there's a catch:

1. The Grid is Local: You can't just move your whole restaurant to a different country. You are bound by local laws (data sovereignty) and the fact that your customers are right here in your city. You can only move your stations to other locations within your region (like moving the salad bar to a nearby town with a wind farm).
2. The Power Changes: The "greenness" of the electricity changes throughout the day. At noon, solar power is high (very green). At night, the grid relies on coal or gas (very dirty).
3. The Rush Hour: If you move your stations too far apart, the waiters (data) take too long to run between them, and your customers get angry because their food is late (latency).

The Problem: Most existing "green" solutions are like moving a massive factory to a different continent. That works for big tech giants (like Google or Amazon) who have global real estate. But for Small and Medium-sized Enterprises (SMEs)—the local restaurants, banks, and shops that make up 99% of businesses—this is impossible. They are stuck in one region, trying to figure out how to be green without slowing down their service or going bankrupt.

Enter: Aceso (The Smart Head Chef)

The paper introduces Aceso, a smart system designed specifically for these local businesses. Think of Aceso as a super-intelligent Head Chef who manages your restaurant's layout in real-time.

Here is how Aceso works, using simple analogies:

1. The "Critical Path" vs. The "Side Dishes"

In your restaurant, some tasks are critical. If the Grill (which cooks the main steak) is slow, the whole meal is late. If the Salad is slow, it's annoying, but the steak is still the priority.

• Aceso's Secret: It realized that you don't need to move everything to be green. You only need to move the "side dishes" (non-critical tasks) that don't hold up the main course.
• The Insight: If you move the Salad Station to a nearby town with a wind farm (green energy), you save carbon. But if you move the Grill too far away, the waiter has to run too far, and the steak gets cold (latency violation). Aceso knows exactly which stations can be moved without ruining the meal.

2. The "Weather Forecast" (Traffic Prediction)

Aceso doesn't just react to the current rush; it looks at the weather forecast.

• It predicts when the lunch rush will hit.
• It knows that at 2:00 PM, the local wind farm is spinning fast (cheap, clean energy).
• The Move: Before the rush even starts, Aceso quietly moves the Salad and Drink stations to the wind-farm town. When the rush hits, they are already there, cooking on clean power.
• The Safety Net: If the rush is too huge, Aceso knows the waiters can't run that far. It quickly moves the stations back to the main kitchen to ensure the food arrives on time.

3. The "Search for the Best Spot" (Optimization)

Imagine trying to find the best seat in a theater with 1,000 seats. If you try every single seat, it takes forever.

• Aceso's Trick: It uses a "pruning" technique. It immediately says, "No, we can't sit in the back row (too far from the stage/latency too high)" and "No, we can't sit in the VIP section (too expensive)."
• By cutting out the bad options instantly, it finds the perfect seat in seconds, not hours. This allows it to make decisions fast enough to keep up with real-time changes in the power grid.

The Results: A Win-Win-Win

When the authors tested Aceso in the real world (using a social network app as their "restaurant"):

• Green: They reduced carbon emissions by 37.4%. That's like taking a huge chunk of the restaurant's pollution and turning it off.
• Cheap: They saved 3.6% on their electricity bill. For a big company, that's millions of dollars; for a small business, it's money they can reinvest.
• Fast: The customers didn't notice a difference. The food (data) still arrived on time, meeting all the "Service Level Objectives" (SLOs).

Why This Matters

For years, the message was: "To be green, you need to be huge and global."
Aceso says: "No. Even if you are a small business stuck in one country, you can still be green, save money, and keep your customers happy."

It's like realizing you don't need to buy a Tesla to be eco-friendly; sometimes, you just need to know exactly when to turn off the lights and when to open the windows. Aceso is the system that tells you exactly when to do that for your digital business.

Technical Summary

Here is a detailed technical summary of the paper "Aceso: Carbon-Aware and Cost-Effective Microservice Placement for Small and Medium-sized Enterprises."

1. Problem Statement

The paper addresses the challenge of deploying microservice applications in a manner that is both carbon-efficient and cost-effective, specifically for Small and Medium-sized Enterprises (SMEs).

• Context: While microservices offer modularity and scalability, their deployment often ignores environmental impact. Data centers contribute significantly to global carbon emissions.
• The Gap: Existing carbon-aware scheduling solutions primarily target batch jobs, high-performance computing (HPC), or serverless workloads. They often assume access to global-scale infrastructure (e.g., AWS Lambda), allowing workloads to shift freely across continents.
• The Constraint: SMEs typically operate within geographically constrained infrastructures (e.g., a single country or the EU) due to data sovereignty regulations (GDPR), privacy laws, and limited resources. They cannot simply "follow the sun" across the globe. Furthermore, microservices are latency-sensitive; moving components across distant regions can violate Service Level Objectives (SLOs) due to increased network latency and complex inter-service dependencies.
• Core Question: How can SMEs deploy microservice applications in a carbon- and cost-efficient manner within geographically limited regions while maintaining strict performance guarantees?

2. Methodology: The Aceso System

Aceso is an adaptive placement system designed to optimize microservice deployment across a limited set of candidate regions. It employs a scalable genetic algorithm (GA) enhanced by insight-driven search space pruning.

System Architecture

Aceso consists of four main components:

1. Traffic Forecaster: Uses a Gradient Boosted Decision Tree (GBDT) model to predict short-term traffic volumes (1-hour horizon) based on historical patterns. This allows proactive rather than reactive placement.
2. Workload Profiler: Characterizes the resource usage (CPU, memory, energy) and latency of each microservice under different traffic loads. It stores this data in a hash table indexed by traffic buckets for fast runtime lookups.
3. Placement Optimizer: The core engine that solves a constrained multi-objective optimization problem. It aims to minimize a weighted sum of carbon emissions and monetary cost while satisfying latency SLOs.
   • Algorithm: A Genetic Algorithm (GA) evolves a population of candidate placements.
   • Search Space Pruning: To handle the exponential complexity of placement decisions, Aceso uses two key insights to prune the search space:
     • Microservice Pinning: It identifies and "pins" (keeps in the base region) microservices that are either legally immovable (e.g., databases due to sovereignty) or performance-critical. Crucially, it pins late-activated microservices and those on the critical path of the dependency graph. Moving these causes severe latency penalties. Only early-activated, non-critical subtrees are considered for offloading.
     • Region Filtering: It uses a Pareto-style rule to discard regions that do not offer a simultaneous improvement in carbon and cost compared to the baseline, preventing unnecessary cost increases.
4. Scheduler: Executes the placement decisions by provisioning resources, migrating services from pre-replicated container images, and rerouting traffic while monitoring SLO compliance.

Optimization Logic

The system triggers re-optimization based on two events:

• Carbon-driven: When predicted carbon intensity in a region deviates significantly (e.g., >20%) from recent trends (e.g., day/night cycles).
• Workload-driven: When forecasted traffic shifts to a new load bucket, indicating a need to scale or re-balance.

3. Key Contributions

1. Characterization of SME Constraints: The paper provides a detailed analysis of why global-scale carbon strategies fail for SMEs and demonstrates that selective offloading of early-activated, non-critical microservice subtrees can reduce emissions without compromising performance.
2. Design of Aceso: A novel system that jointly optimizes carbon, cost, and latency using a GA-based approach with insight-driven pruning. This makes the optimization tractable for real-time decision-making.
3. Comprehensive Evaluation: Extensive validation using real-world deployments (DeathStarBench social network app on AWS) and high-fidelity simulations.
4. Open Source: The authors commit to open-sourcing Aceso and a high-fidelity simulator to facilitate community adoption.

4. Experimental Results

The evaluation was conducted on a real-world deployment (Spain vs. Sweden) and a large-scale simulation (100 microservices across EU regions).

• Carbon Reduction:
  • In real-world experiments, Aceso reduced carbon emissions by 37.4% on average compared to a static single-region deployment.
  • In North American scenarios, it achieved a 21.2% reduction.
• Cost Efficiency:
  • Aceso reduced operational costs by 3.6% on average.
  • It demonstrated that greener regions are not necessarily more expensive; in some cases (e.g., Stockholm), they were cheaper.
• Latency & SLO Compliance:
  • Aceso maintained strict SLOs (300ms end-to-end latency).
  • In real-world tests, it achieved an average latency of 93.8ms (EU) and 40.1ms (NA) with <0.2% SLO violations.
  • Crucially, it avoided the severe latency penalties seen in naive strategies that move late-activated services.
• Performance vs. Baselines:
  • Compared to Linear Programming (LP) (which finds optimal solutions but takes hours), Aceso converges in seconds (25.8s).
  • Compared to other heuristics (Caribou, Vanilla GA), Aceso offers better stability (fewer unnecessary migrations) and lower latency while achieving near-optimal carbon savings.
• Scalability: Aceso scales effectively to 1,000 microservices and global region sets, solving placements in under 2.2 minutes, whereas other approaches fail to scale or take impractical amounts of time.

5. Significance

• Democratizing Green Cloud: Aceso bridges the gap between high-level carbon-aware research and the reality of SMEs. It proves that sustainable computing is achievable even without global infrastructure or massive budgets.
• Latency-Aware Sustainability: It challenges the notion that carbon reduction requires sacrificing performance. By leveraging the activation order of microservices, it shows that strategic, partial offloading is more effective than blanket migration.
• Practical Viability: The system is designed for real-world constraints (data sovereignty, limited regions) and integrates with standard cloud tools (Kubernetes, AWS), making it a viable solution for immediate adoption.
• Economic Incentive: By demonstrating that carbon reduction can also lead to cost savings, Aceso provides a strong business case for SMEs to adopt green practices without financial penalty.

In summary, Aceso represents a significant step forward in making cloud computing sustainable for the "real-world majority" of businesses, balancing environmental goals with the strict performance and regulatory requirements of modern microservice architectures.