Your community.
On your jurisdiction. Federated by default.

Email deliverability. The Mastodon application itself is well-engineered, the Docker Compose stack assembles in an afternoon, and Postgres tuning is a known science. The part that breaks people's instances is signup confirmations, password resets, and federation moderation notifications failing to reach Gmail / Outlook365 inboxes because the SMTP path from a fresh VPS IP is treated as guilty-until-proven-innocent. Plan for an external transactional mail relay (Mailgun, Postmark, Amazon SES) from day one — it costs $10/mo and removes 80% of the operational pain.

Mastodon's Postgres database grows roughly proportional to the union of "what your local users post" plus "everything from every account anyone here has ever followed on every other instance". For a 100-active-user instance with healthy outward federation, expect 20–40 GB of database after the first year and 80–150 GB after three years. Media (uploaded images / videos / boosts) lives separately and is the bigger storage problem — see the next question.

Yes, and you should. Mastodon supports any S3-API-compatible backend via the S3_ENABLED env var. NordBastion does not currently operate first-party object storage, so the pragmatic choice is an external provider — Backblaze B2 is the standard recommendation for self-hosted instances (predictable pricing, no egress fees when paired with Cloudflare in front). With offloaded media a Ravelin's 480 GB NVMe holds Postgres and Sidekiq job queues for years; without offload, the same 480 GB fills in 12–18 months on a moderately-active instance.

Ravelin (8 vCPU, 16 GB, 480 GB) is comfortable up to ~100 actively-posting daily users with healthy outward federation. The bottleneck that breaks first is Sidekiq workers under inbound federation bursts (a viral post boosted by a 50k-user instance produces a transient queue spike that wants more workers, which wants more RAM). At ~300 active users or once Sidekiq processed-jobs/sec sits >100 sustained, the Bulwark tier earns its keep — more cores, more RAM headroom, and the freedom to dedicate one core to Elasticsearch if you want search.

Defederation तब होता है जब आपका इंस्टेंस (या कोई अन्य इंस्टेंस) किसी peer से संदेश स्वीकार करना बंद कर देता है — व्यवहार में उन इंस्टेंस को ब्लॉक करने के लिए उपयोग किया जाता है जो आपके समुदाय के मानकों का उल्लंघन करने वाली सामग्री होस्ट करते हैं। Mastodon एडमिन UI प्रति-इंस्टेंस suspend/silence/limit टॉगल उजागर करता है; सामाजिक लागत वास्तविक है (आपके उपयोगकर्ता ब्लॉक किए गए इंस्टेंस पर लोगों के साथ थ्रेड खो देते हैं) और मॉडरेशन लागत वास्तविक है (किसी को निर्णय लेना होगा)। अपना खुद का इंस्टेंस चलाने का उद्देश्य यह है कि यह आपका निर्णय है — आपके समुदाय के मानकों के आधार पर, किसी कॉर्पोरेट trust-and-safety पाइपलाइन के नहीं।

For an instance under ~50 users with curated signups (invite-only or admin-approved), moderation is a few minutes per week — closing spam-account reports, occasionally limiting a peer instance. For open-registration instances the effort scales superlinearly with signup volume because the spam-bot economy treats every open ActivityPub endpoint as a target. The standard defence is approval-required registrations + a clear application question — drops automated signups by 99% with negligible friction for real humans.

Yes, for a specific and underrated reason: as the admin you receive moderation reports, takedown requests, and the occasional GDPR data-subject request. Sweden, Finland, Norway and Iceland have a well-defined legal framework for what a hoster and a small forum-operator owe in response — and crucially, no equivalent of the US DMCA notice-and-staydown machinery that lets a single accusation force pre-emptive takedown. You still moderate; you just do it on your terms rather than under threat of a hoster auto-acting.

Hosting-wise: Mastodon is heaviest (Ruby on Rails + Sidekiq + Postgres + Redis + optional Elasticsearch), Pleroma/Akkoma are lighter (Elixir, single OTP release), Sharkey/Misskey-family run on Node + Postgres. A Ravelin handles any of them comfortably for ~100 users; a Sentinel can run Pleroma/Akkoma for personal-instance use where the same tier would not be enough for Mastodon. Pick on community-fit (which features matter, which UI your people prefer), not on resource curve — the difference is small at this scale.