Skip to main content
Archil disks use a high-speed, durable cache to enhance performance when working with data from data sources. You are only charged for data actively stored in this cache layer. For pricing details, see Pricing & Billing.

​
File data

When reading data from a data source, Archil automatically caches it in the high-speed storage layer. Cached data remains active for up to 1 hour after your last access. Writes are buffered in the high-speed storage layer for coalescing before being flushed to the data source. Recently written data remains active for up to 1 hour after it has been flushed. The amount of active data may not exactly match the amount you’ve read or written. Archil prefetches and caches data ahead of sequential reads, which can result in additional cached data. Rapid overwrites to the same file will also result in additional active data while Archil asynchronously coalesces and flushes the writes.

​
File metadata

Metadata accesses (e.g., find, ls, stat) do not cause file data to become active and incur no metering charges. Metadata writes (e.g., rename, chmod) are counted as 32 KiB of active data per operation until the change is propagated to the data source. Taking ownership of a path with archil checkout lets Archil coalesce the many small attribute updates a workload produces into larger chunks before metering them. For example, git clone-ing a 15 MB repository into a checked-out directory is metered as roughly fifteen 1 MiB writes rather than the thousands of individual metadata operations the client issues.

​
How billing is calculated

Archil meters the amount of active data each minute. Your monthly bill is the time-weighted average of active data over the billing period, charged at $0.20 per GiB-month. For example, if you store 100 GiB for half of a month, you are billed for 50 GiB-months. You can monitor your active data usage in the Archil console on each disk’s Usage tab, or on the Billing page for an account-wide view.

​
Branches and checkpoints

Branches and checkpoints are copy-on-write. Forking a branch or taking a checkpoint does not duplicate the underlying data — you are billed once for the shared base data, plus only the unique data each branch or checkpoint adds. For example, a 100 GiB parent disk with 50 branches that each modify ~5 GiB of unique data is metered as roughly 350 GiB total — the shared 100 GiB base plus the ~250 GiB of unique per-branch changes — not as 51 full copies (~5,100 GiB). A checkpoint retains its unique data for as long as it exists; deleting a checkpoint removes that data from your bill.