How Erasure Coding Cuts Storage Costs in Ozone: A Deep Dive

01 Introduction

Erasure Coding (EC) is an encoding technique used in RAID and communication to ensure data reliability. EC‑encoded files can be decoded even when a small portion is damaged. Ozone originally guarantees reliability with multiple replicas, typically three copies, which multiplies storage cost. To lower this cost, Ozone adopts EC.

1.2 Data Hot‑Cold Characteristics

Data is classified by access frequency: hot data is frequently accessed and stored on high‑performance media; cold data is rarely accessed and stored on cheaper media. IDC reports that in production environments hot data accounts for about 10%, cold data 60%, and the remaining 30% is warm data.

02 EC in Ozone

2.1 Advantages of EC

Using three‑replica storage, a 2 MB file consumes 6 MB of raw capacity. With EC using a 3‑2‑512 KB configuration, the same file occupies 4 MB, saving roughly 33 % of storage. A 4‑2‑512 KB configuration reduces the occupied space to 3 MB, saving about 50 %. Higher EC ratios yield greater storage savings.

2.2 EC in Object Storage Products

2.2.1 Intelligent Tiering

Hot files, critical for business, are stored with three replicas for fast access. Warm or cold files, which tolerate higher latency, are stored using EC encoding, reducing cost. Based on file lifecycle, Ozone offers an intelligent tiered storage product that automatically migrates data from replica‑based storage to EC‑based storage as access frequency declines.

2.2.2 Archive Storage

When data becomes cold, Ozone provides an archive storage tier that uses high EC ratios. Archive storage is suitable for long‑term, infrequently accessed data, but incurs higher read latency and requires a thawing process that temporarily creates a three‑replica copy for reading.

2.3 Dynamic EC Ratios

Different EC ratios affect storage savings, and block size also influences the final occupied space. For a 512 KB file with a 3‑2 EC ratio, using block sizes of 256 KB, 512 KB, and 1 MB results in total storage consumption of 1 MB, 1.5 MB, and 2.5 MB respectively. Selecting the EC configuration that yields the smallest space usage maximizes storage efficiency.

2.4 Limitations of EC

Performance impact: with a 6‑3‑512 KB EC configuration, a file block is split into nine blocks distributed across different nodes, requiring at least six DataNodes to respond for a read, increasing request load compared to three‑replica mode.

Repair cost: damaged blocks need CPU‑intensive reconstruction, and higher EC ratios increase the probability of block loss.

03 EC Repair Throttling

Higher EC ratios raise the likelihood of block loss, triggering two repair mechanisms: read‑triggered asynchronous repair when a missing replica is accessed, and periodic reports from DataNodes to the Storage Container Manager (SCM) that initiate repair if a replica is missing.

Repair tasks consume CPU and, without limits, can overload a DataNode. Ozone introduces a token‑bucket based throttling mechanism that caps the number of concurrent repair tasks per DataNode. Tasks exceeding the limit are queued and may be dropped if they wait beyond a maximum timeout.