This approach "flattens" exponential differences, ensuring that no single resource is overwhelmed by extreme outliers.
| Feature | Linear Loadshare | Log10 Loadshare | | :--- | :--- | :--- | | | Fails; large requests dominate queues | Treats large requests as incrementally larger, not infinitely so | | Dynamic Range | Effective for 1:1000 ratios | Effective for 1:1,000,000,000 ratios | | Small File Overhead | High connection overhead for small files | Groups small files into lower-order buckets | | Cache Utilization | Poor; large files evict small ones | Balanced; cache is stratified by size class | log10 loadshare
Because in large-scale systems, capacity often spans orders of magnitude. This approach "flattens" exponential differences
def select_server(servers, file_size): bucket = get_log10_bucket(file_size) required_share = bucket_weight[bucket] # Sort servers by current load (in log10 units) eligible = sorted(servers, key=lambda s: s.current_log_load) # Assign to least loaded server eligible[0].current_log_load += required_share return eligible[0].id log10 loadshare