Infrastructure Overview
This page provides a high-level overview of ABI's computing and network infrastructure. For user-facing documentation, see Cluster Basics.
Architecture
ABI's infrastructure runs on FreeBSD physical hosts with compute workloads running in bhyve VMs. Services are isolated using FreeBSD jails.
┌──────────────┐
│ Internet │
└──────┬───────┘
│
┌──────┴───────┐
│ Firewall │
│ / Router │
└──────┬───────┘
│
┌───────────────────────────┼──────────────────────────┐
│ │ │
┌─────┴──────────────┐ ┌────────┴─────────────┐ ┌────────┴─────────────┐
│ GEONOSIS (nas0) │ │ BANE │ │ MUSTAFAR (nas1) │
│ FreeBSD │ │ FreeBSD │ │ FreeBSD │
│ │ │ │ │ │
│ ZFS: genomic data │ │ Jails: │ │ ZFS: home, user, │
│ /mnt/nas0/proj │ │ - nginx │ │ proj, db │
│ /mnt/nas0/user │ │ - LDAP │ │ /mnt/home │
│ │ │ - Forgejo (git) │ │ /mnt/nas1/proj │
│ bhyve VMs: │ │ - PostgreSQL │ │ /mnt/nas1/db │
│ - ssh-01 (2C/4G) │ │ - MySQL │ │ │
│ - ssh-02 (4C/8G) │ │ - DNS │ └──────────────────────┘
│ - thin-01(64C/384G) │ - DHCP │
│ - thin-02(64C/384G) │ │ ┌───────────────────────┐
│ - thick-01(64C/768G) │ Devuan VM on bhyve: │ │ HOTH (bak1) │
│ - dl-01 (2C/8G) │ │ - slurmctld │ │ FreeBSD │
│ - dl-02 (2C/8G) │ │ │ │ │
│ - rshiny0 (4C/16G) └──────────────────────┘ │ ZFS backups via │
│ - rshiny1 (4C/16G) │ zelta (send/recv) │
│ - (+ legacy VMs) │ │ homes + some projects│
└────────────────────┘ └───────────────────────┘
Physical Hosts
ABI runs on 4 physical FreeBSD servers:
| Hostname | Alias | Role | OS | Key Function |
| geonosis | nas0, genomic | HPC hypervisor + NAS | FreeBSD | 2x AMD EPYC 7702 (128C/256T). Runs all bhyve VMs. Hosts genomic data ZFS pool. |
| mustafar | nas1 | Primary NAS | FreeBSD | ZFS storage for /mnt/home, /mnt/nas1/proj, /mnt/nas1/db. Serves NFS to all VMs. |
| bane | – | IT services server | FreeBSD | Runs all infrastructure services in jails (LDAP, nginx, git, DNS, DHCP, databases). Hosts slurmctld in a Devuan VM. |
| hoth | bak1 | Backup server | FreeBSD | ZFS backup target using zelta (ZFS send/recv). Backs up homes and selected projects. |
Virtual Machines
All HPC VMs run on geonosis using bhyve. The Slurm controller runs in a Devuan VM on bane.
User-facing VMs (on geonosis)
| VM | vCPUs | RAM | OS | Slurm Partition | Purpose |
| ssh-01 | 2 | 4G | *TODO* | N/A | Login node (ssh.abi.am) |
| ssh-02 | 4 | 8G | *TODO* | N/A | Login node (ssh.abi.am) |
| thin-01 | 64 | 384G | *TODO: Devuan/Ubuntu?* | compute (default), thin | General computation |
| thin-02 | 64 | 384G | *TODO* | compute (default), thin | General computation |
| thick-01 | 64 | 768G | *TODO* | compute (default), thick | High-memory computation |
| dl-01 | 2 | 8G | *TODO* | download | Data downloads only |
| dl-02 | 2 | 8G | *TODO* | download | Data downloads only |
| rshiny0 | 4 | 16G | *TODO* | N/A | R Shiny server |
| rshiny1 | 4 | 16G | *TODO* | N/A | R Shiny server |
Infrastructure VMs
| VM | Host | vCPUs | RAM | OS | Purpose |
| slurmctld VM | bane | *TODO* | *TODO* | Devuan | Slurm controller daemon (slurmctld) |
Legacy / testing VMs (on geonosis)
These VMs are for internal IT use. *TODO: Document when needed.*
| VM | vCPUs | RAM | Notes |
| comp0 | 32 | 32G | Stopped – legacy compute node |
| dna0 | 8 | 32G | Stopped – *TODO* |
| mitte-dev-01 | 2 | 4G | Stopped – development/testing |
Services (on bane)
All infrastructure services run in FreeBSD jails on bane (managed via jailer), providing isolation and easy management.
Public-facing jails
| Jail | Hostname | IPv4 | Service |
| www-01 | www-01.abi.am | 37.26.174.181 | nginx (web server, reverse proxy) |
| git-01 | git-01.abi.am | 37.26.174.182 | Forgejo (Git hosting) |
| mx-01 | mx-01.abi.am | 37.26.174.190 | Mail server |
Internal jails
| Jail | Hostname | Network | Service |
| ldap-01 | ldap-01.local.abi.am | 172.20.42.0/24 | LDAP (slapd). See LDAP |
| dns-01 | dns-01.local.abi.am | 172.20.42.0/24, 172.20.200.0/24 | DNS |
| dhcp-01 | dhcp-01.local.abi.am | 172.20.42.0/24, 172.20.200.0/24 | DHCP |
| psql-01 | psql-01.local.abi.am | 172.20.200.0/24 | PostgreSQL |
| mysql-01 | mysql-01.local.abi.am | 172.20.200.0/24 | MySQL |
| adg-01 | adg-01.local.abi.am | 172.20.200.0/24 | AdGuard (DNS filtering) |
| nms-01 | nms-01.local.abi.am | 172.20.200.0/24, 172.20.42.0/24 | Network monitoring |
| unifi-01 | unifi-01.local.abi.am | 172.20.200.0/24 | UniFi controller (WiFi) |
| wiki-01 | wiki-01.local.abi.am | 172.20.200.0/24 | DokuWiki |
Infrastructure VM
| VM | Host | OS | Service |
| slurmctld | bane | Devuan | Slurm controller (port 6817) |
For the full jail list including stopped jails, see Server Inventory: bane.
Network
See Network for detailed network documentation.
Summary:
Public subnet: 37.26.174.176/28 – internet-facing services (www-01, git-01, mx-01, ssh.abi.am)
Internal subnet 1: 172.20.42.0/24 – core infrastructure services (LDAP,
DNS, DHCP)
Internal subnet 2: 172.20.200.0/24 – internal services (databases, monitoring, wiki, UniFi, AdGuard)
DNS: internal
DNS served by
dns-01 jail on bane
DHCP: served by dhcp-01 jail on bane
WiFi management: UniFi controller on unifi-01 jail
*TODO: VLAN segmentation details, switch configuration, inter-subnet routing.*
Storage Architecture
ABI uses two ZFS-based NAS servers with storage served over NFS. ZFS provides transparent compression, snapshots, and data integrity.
mustafar (nas1) -- Primary user storage
| Mount Point | ZFS Dataset | Capacity | Purpose | Quota |
/mnt/home | nas1:/znas1/abi/home | ~32 TB pool | User home directories | ~12G per user |
/mnt/nas1/proj | nas1:/znas1/abi/proj | ~32 TB pool | Shared project data | Per-project |
/mnt/nas1/db | nas1:/znas1/abi/collections/db | ~32 TB | Reference genomes & databases | Read-only for users |
geonosis (nas0) -- Genomic data + user workspaces
| Mount Point | ZFS Dataset | Capacity | Purpose | Quota |
/mnt/nas0/user | nas0:/znas0/abi/user *TODO: confirm dataset path* | *TODO* | Personal user workspaces | ~100G per user |
/mnt/nas0/proj | *TODO: dataset path* | *TODO* | Additional project storage | Per-project |
hoth (bak1) -- Backups
| What | Backed up? | Method |
User home directories (/mnt/home) | Yes | ZFS send/recv via zelta |
| Some project directories | Selected projects only | ZFS send/recv via zelta |
Reference databases (/mnt/nas1/db) | *TODO* | *TODO* |
Slurm Architecture
| Component | Where | Notes |
| slurmctld (controller) | Devuan VM on bane | Manages the job queue and scheduling |
| slurmd (compute daemon) | thin-01, thin-02, thick-01, dl-01, dl-02 | Runs on each compute/download VM |
| Configuration | *TODO: path to slurm.conf* | *TODO* |
| Slurm version | *TODO* | |
Operating Systems
| Role | Machine | OS |
| Physical hosts | geonosis, mustafar, bane, hoth | FreeBSD *TODO: version* |
| Login VMs | ssh-01, ssh-02 | *TODO: Devuan? Ubuntu?* |
| Compute VMs | thin-01, thin-02, thick-01 | *TODO: Devuan? Ubuntu?* |
| Download VMs | dl-01, dl-02 | *TODO* |
| R Shiny VMs | rshiny0, rshiny1 | *TODO* |
| Slurm controller VM | slurmctld (on bane) | Devuan |
Key Design Decisions
FreeBSD everywhere for physical hosts – provides ZFS, bhyve, and jails natively.
bhyve for virtualization – all HPC workloads run in VMs on geonosis, providing isolation and resource control.
FreeBSD jails for services on bane – lightweight isolation for LDAP, web, git, databases,
DNS, DHCP.
ZFS for all storage – transparent compression, snapshots, send/recv for backups.
zelta for backups – ZFS replication from mustafar/geonosis to hoth.
No environment modules – all bioinformatics software installed globally on compute VMs.
Related Pages
-
-
Network – Network topology and configuration
-
-
-
-
