Memory for the working set, NVMe for everything else, and cores allocated to the database when its load is steady. On the AMD EPYC 4584PX with DDR5 memory on 2 channels, in Mumbai, Singapore and Limburg.
Where databases start
Prices exclude GST and are the same in every region.
| Plan | Memory | PostgreSQL shared_buffers | MySQL buffer pool | NVMe |
|---|---|---|---|---|
| VPS Mini | 4 GB | 1 GB | 2 GB | 40 GB |
| VDS MicroCore | 8 GB | 2 GB | 6 GB | 100 GB |
| VDS FusionCore | 16 GB | 4 GB | 12 GB | 200 GB |
| VPS Elite | 16 GB | 4 GB | 12 GB | 180 GB |
| VDS OmniCore | 24 GB | 6 GB | 18 GB | 500 GB |
| VDS PrimeCore | 32 GB | 8 GB | 24 GB | 800 GB |
| VDS TitanCore | 48 GB | 12 GB | 36 GB | 1 TB |
| VDS UltraCore | 64 GB | 16 GB | 48 GB | 1.5 TB |
Starting values for a server that runs only the database, from each plan's memory (the operating system reports a little less). PostgreSQL: 25% of memory, from the PostgreSQL 18 manual, which adds that more than 40% is unlikely to help, since it also relies on the operating system's cache. MySQL: what innodb_dedicated_server sets, half of memory up to 4 GB and three quarters above it, from the MySQL 8.4 manual. Tune from there with your own workload.
A database behind a small site or an internal tool spends most of its time waiting for the next query. A vCPU is processor time on shared cores, which suits that. An API, a staging copy of production, or a small database.
Under steady load a database keeps its cores busy all day. On a VDS they are allocated to your server and not scheduled for anyone else, and every core comes with 4 GB of memory. A production database with steady load, or a large game server.
Have the database listen on localhost and add no rule for its port. With Default for unlisted ports set to Deny all, a port without a rule is dropped before it reaches the server.
Add a rule for the database port, then allow only the application server's address in the server's own firewall. Source rules in the panel apply to the whole server, not to one port, so the per-port limit belongs on the server. A firewall on Linux
Traffic passes 17 Tbps of upstream mitigation capacity, then Prism on the host, whichever port it is aimed at. Your rules decide what is reachable at all.
pg_dump, mysqldump or your engine's own tool on a schedule, copied to storage that is not on the same server, so one failure cannot take both.
The Backups tab lists the snapshots taken on your plan's schedule, if it includes one. Support restores them on request; nothing in the panel undoes a restore.
A backup you have never restored is a guess. Restore one onto a spare server now and then, and time how long it takes.
Cores and traffic change while the server keeps running. FusionCore to OmniCore, as an example.
Memory and NVMe take effect when the server boots again. Restart from the panel when it suits you.
No. You get a server with Full root / Administrator access and install the database yourself, so you choose the engine, the version and every setting. You are responsible for its security and its backups, as the Terms of Service set out.
Anything that runs on Linux, FreeBSD or Windows Server: PostgreSQL, MySQL, MariaDB, MongoDB, Redis, SQL Server and the rest. The server is yours to install on.
A small or quiet database is fine on a VPS. When the load is steady, a VDS allocates whole cores to it, which are not scheduled for anyone else, and gives 4 GB of memory per core against 2 GB per vCPU on a VPS.
If the application runs on the same server, have the database listen on localhost and do not add a rule for its port. If the application runs elsewhere, add a rule for the port and allow only the application server's address in the server's own firewall, since source rules in the panel apply to the whole server rather than one port.
Scheduled snapshots are included only where your plan says so, and support restores them on request. A snapshot of a running database is no substitute for a dump: take your own, copy them off the server, and test that they restore.
From 20 GB of NVMe on the smallest VPS to 1.5 TB on UltraCore. Beyond that, bare metal is built with the drives you need and quoted per build.
Yes. When you move to a larger plan, CPU and traffic change straight away, while memory and disk take effect at the next restart. Restart from the panel when a short stop of the database suits you.
The same one as the application that queries it, since every query crosses the distance between them. Hardware, protection and prices are the same in Mumbai, Singapore and Limburg.
4 allocated cores, 16 GB of DDR5 and 200 GB of NVMe, with Prism in front of it.
