Architecture / SLA

SLA

LAST UPDATED IN SEPTEMBER 2026

The service level agreement for the MELAPS clusters. The signed reservation agreement takes precedence where it differs.

Availability target

Reserved capacity is available 99.5% of the minutes in a calendar month, measured per reservation. A minute counts as unavailable when a reserved GPU cannot start a job for a MELAPS reason: hardware failure, host fault, network fault, or unannounced maintenance.

Announced maintenance windows do not count against availability: announced at least 7 days ahead, lasting at most 4 hours, and occurring at most twice a month. On demand and batch carry no availability commitment: capacity is listed when it is free. A queued job is not an outage.

Credits

Credits are the only remedy and are applied automatically, without a claim, to the next invoice. Where a reservation is unavailable for more than 24 hours, the customer may cancel the reservation without penalty.

Hardware

A GPU that reports uncorrectable ECC errors, falls off the bus, or fails a health check is replaced within 4 hours. If a whole cluster fails, it is rebuilt within 36 hours. All recovery times are counted from when MELAPS detects the problem, not from when it is reported.

Data

Data stored on the cluster is not guaranteed to survive a node failure; a failed node can lose data. There is no automatic backup — you are responsible for your own backups. Data stored outside the cluster is not affected by a node failure.

Security

Access to the cluster is limited to registered users. Data is encrypted in transit, not encrypted at rest. API keys can be revoked at any time. If you suspect a key leaked, contact support.

Support

Support is available 24/7 by email at support@melaps.system. First response: 4 hours for critical issues, 24 hours for non-critical issues.

Changes

This page can change without notice. Changes apply to the following month, not the current month. Price changes do not apply to active reservations.