To help provide the best possible pricing, we deploy servers at a scale and operational model that prioritizes cost efficiency for our customers.
In the event of a server failure, you may request a replacement server. Please note that storage is not transferred to replacement servers—see also: Backups. For storage servers, we can replace a failed drive during our standard maintenance cycle (every 1–2 months) upon request. If a faster resolution is required, we recommend requesting a replacement server instead—see also: Hot Spares.
The best parallel to this behavior are the lower-cost “Spot” instances at AWS, which have a lower price-point, but can be shut down as circumstances change.
As described prior, these servers are designed for distributed availability, so systems that support individual node failures are best. You can achieve greater than 5-9 SLA through through your software design, by setting up your service crossing multiple zones.
We recommend designing workloads to tolerate full node failures, such as ephemeral services or distributed application architectures. Where available, you may also request servers in separate pods within the same site for additional redundancy.
Full details on our SLA are in the Service Level Agreement section of our Terms and Policies.
For security and privacy reasons, we do not maintain backups of customer servers. You are responsible for managing backups for your systems.
All storage servers include built-in hot spares in line with our fail-in-place policy. The number of spares varies by configuration: flash storage systems include one spare volume, while HDD-based systems include either two or four spare drives, depending on configuration (36- or 72-drive systems).
These spare drives should not be included in your active volume configurations, except as designated spares (i.e., not as part of a RAID group). If a drive in your RAID group fails, you may replace it with a hot spare and notify us. We will then schedule a replacement during the next maintenance cycle.
For large HDD based arrays, we recommend multiple RAID6 volumes no larger than 8-10 drives, joined in a concat by LVM (or stripe, we haven’t tested the diff performance characteristics, just make sure not to over-fragment with stripe boundaries). That, or ZFS RAIDZ3. In both cases reserve the correct amount of hot spares. Feel free to reach out if further clarity is needed.
