ARCHIL
All comparisons

Archil vs Mountpoint for Amazon S3

Pick Mountpoint for free, open-source, high-throughput reads of large S3 objects and sequential new-file writes, such as training, data lakes and ETL on AWS. Pick Archil for full POSIX on that data (in-place edits, renames, links, git, npm, SQLite), coordinated writers, millisecond branches and buckets beyond S3.

Where Archil stands out

  • Edit files in place

    Random writes, appends, directory renames and links work, so git, npm and SQLite run on the disk.

  • Many coordinated writers

    Up to 5,000 clients per disk with checkouts; Mountpoint doesn't coordinate writers.

  • More than Amazon S3

    Disks also sync with GCS, R2 and S3-compatible stores.

Archil and Mountpoint for Amazon S3, side by side

What it is
Archil: Managed POSIX file system over your bucket or Archil-managed storage, with an S3 API and sandboxes
Mountpoint for Amazon S3: Open-source FUSE client that turns file operations into S3 calls
Data in the bucket
Archil: Ordinary objects at their paths
Mountpoint for Amazon S3: Ordinary objects at their paths
Edit existing files
Archil: Yes, including random writes and appends
Mountpoint for Amazon S3: Sequential writes only; overwrite is opt-in and replaces the whole file
Rename
Archil: Atomic, including directories
Mountpoint for Amazon S3: Files on S3 Express One Zone only; no directory rename
Symbolic and hard links
Archil: Yes
Mountpoint for Amazon S3: No
Several machines writing
Archil: Yes, via delegations or conditional mode
Mountpoint for Amazon S3: No coordination between writers to one object
Shared cache
Archil: SSD storage layer shared by every client
Mountpoint for Amazon S3: Local per client; optional S3 Express One Zone cache for objects up to 1 MiB
Snapshots and branches
Archil: Whole-disk checkpoint p50 13 ms, branch p50 33 ms
Mountpoint for Amazon S3: None; S3 Versioning recommended against accidental deletes
Supported buckets
Archil: S3, GCS, R2, S3-compatible
Mountpoint for Amazon S3: Amazon S3 only; other S3-compatible stores unsupported
Cost model
Archil: Active storage per GiB-month, plus your bucket
Mountpoint for Amazon S3: Open source; you pay for S3 requests and storage

Measured on one EC2 host

MethodOne AWS c7gn.2xlarge (arm64) in us-east-1 running two independent mounts of each file system. Archil client 0.9 with a disk in aws-us-east-1; Amazon S3 Files through amazon-efs-utils 3.0.1 (NFS 4.2), with a mount target in the same Availability Zone and a versioned bucket; Mountpoint for Amazon S3 1.22.3 with --allow-overwrite and --allow-delete. Medians of 10 to 200 operations per row; new-file-system rows are medians of 3 to 5 trials. Method.

New file system to first durable write

Lower is better
  • Archil0.34 s
  • Mountpoint for Amazon S31.50 s

    new bucket

Mount an existing file system

Lower is better
  • Archil44 ms
  • Mountpoint for Amazon S3299 ms

Create a file, write 4 KiB, fsync, close

Lower is better
  • Archil4.2 ms
  • Mountpoint for Amazon S3244 ms

Second client opens and reads that file

Lower is better
  • Archil1.2 ms
  • Mountpoint for Amazon S356 ms

Second client sees a newly created file

Lower is better
  • Archil30 s

    default client cache expiry

  • Mountpoint for Amazon S3284 ms

    no metadata cache by default

Read 1 GiB sequentially

Higher is better
  • Archil902 MB/s
  • Mountpoint for Amazon S3585 MB/s

How they differ

Architecture

Mountpoint is one client with nothing else to run; it infers directories from / in keys. Its CSI driver binds a bucket to a Kubernetes volume.

Archil puts a managed SSD layer, replicated across Availability Zones, between client and bucket and syncs in the background. That layer enables directory renames, in-place edits and delegations; disks mount in Kubernetes through the CSI driver.

File system semantics

Mountpoint supports only what maps onto S3: a file can't be read while being written, and locks, chmod, chown and extended attributes aren't supported. Its README says it isn't meant for editing files, such as a Git repository or vim.

Archil supports standard POSIX operations, including modes, ownership and flock/fcntl locks within a client.

Data durability and consistency

Mountpoint makes an object visible to S3 clients on close or fsync, strongly consistent by default (caching relaxes this to the metadata TTL), with S3's durability.

Archil makes a write durable on fsync, designed for 99.999% durability across Availability Zones before it reaches your bucket, and its S3 API sees it immediately. Bucket sync is asynchronous, so direct bucket readers can briefly see an older version. If they must see each file on close, Mountpoint wins.

Sharing and branching

Many Mountpoint clients can read one bucket, which suits fanning out training; each file has one writer.

Archil shared mode checks out the paths each client writes; branches give each agent or experiment its own copy. Branching disks synced to your own bucket is an Enterprise feature; see plan features.

Pricing model

Mountpoint is Apache 2.0; you pay S3 storage and requests, plus Express One Zone if you use the shared cache. See Amazon S3 pricing.

Archil serves active data from its own layer, so repeated reads aren't each an S3 request. See Archil pricing.

Which one should you use?

Use Mountpoint for Amazon S3 when

  • reads dominate, over large objects such as training data or data lakes
  • one client at a time writes new files sequentially
  • you want a free, open-source client on AWS with no service in the data path
  • S3 readers must see each file as soon as it's closed

Use Archil when

  • tools edit in place, rename directories, use links, or run git, npm or SQLite
  • several machines write the same tree
  • each agent or experiment needs its own branch in milliseconds
  • data is in GCS, R2 or another S3-compatible store, or compute is outside AWS
  • agents need a sandbox next to the data

Questions people ask

More comparisons

Or see every comparison, grouped by sandboxes, file systems and object storage.