Architecture

Archil disks provide a POSIX-compatible file system backed by object storage. In managed mode, the backing bucket is managed by Archil. In bring your own bucket (BYOB) mode, the bucket belongs to you. In both modes, active data is temporarily held in Archil’s high-performance storage layer, and newly committed writes are durable there while they synchronize to the backing bucket. With Archil’s custom file protocol, developers can achieve high-speed access to shared data, even if it requires lots of small files.

​

Core components

The Archil storage layer runs in Archil-managed cloud accounts, or in your own environment with BYOC on AWS, Google Cloud, Azure, or bare metal. The backing bucket belongs to Archil in managed mode and to you in BYOB mode.

​

Archil file protocol

The Archil client implements a new, custom storage protocol that enables full POSIX-compliance and high-performance for small-file workloads, such as git. The Archil client supports all Linux distributions. For more information, see the reference documentation for the Archil client.

​

Archil storage system

Archil disks are hosted on a shared cloud service, the Archil storage system. This centralized, durable cache stores frequently accessed data and recently written data. The cache provides near-infinite storage that automatically expands to the working set size of your application. The Archil storage service is designed to provide sub-millisecond latency for data access. Because the Archil cache is shared across all clients, you can achieve low-latency access to reference data from multiple servers simultaneously. The Archil storage system efficiently translates POSIX-like requests into other protocols, such as Amazon S3, even when those operations (such as rename) are not supported by the underlying data source. Archil disks are always strongly read-after-write consistent for reads that reach the server, including the Archil S3 API. Writes are durable as soon as fsync() returns. Mounted clients may continue to serve cached data until their caches refresh; see Consistency. Archil also copies writes to the backing bucket in the background. Clients that read that bucket directly see a separate synchronization delay.

​

Security

Archil encrypts all data at rest (AES-256) and in transit (TLS 1.2 or higher), with access control via IAM roles and scoped API keys. Archil is SOC 2 certified and HIPAA/GDPR compliant. For full details, see Security & Compliance.

​

Data durability

Archil is designed for 99.999% (5 nines) durability for newly written data before it is synchronized to object storage, and 99.99% (4 nines) availability. We can sign an availability SLA for Enterprise customers; contact us to discuss it. Archil’s caching layer automatically stores data redundantly within multiple Availability Zones in the disk’s region. A successful fsync() guarantees that the written data has been committed redundantly across these Availability Zones; the 5-nines durability applies from the moment fsync() returns, before the data has synchronized to your data source. There is no single point of failure that can cause data loss. An Archil disk is not a single server or cache node — if an individual node, or even an entire Availability Zone, fails before your data has synchronized to the data source, the disk continues serving your data from the remaining Availability Zones and synchronization completes normally. No data is lost, and no action is required on your part.