Every server in ColoTycoon is assembled by hand from a barebone chassis and separate parts. Pre-built machines are no longer sold new, so the parts you choose decide how many customers a box can carry, how much power it draws and what happens when a component dies.

The assembly flow

A chassis is a case with a motherboard, slots and bays, but no processor, memory, disks or power supply. You buy it from the vendor shop or the second-hand market, open it on the Nodes page and seat parts into their slots.

  1. Seat at least one processor in a socket.
  2. Add at least one memory module.
  3. Add at least one disk, then choose a storage layout.
  4. Fit at least one power supply with enough watts for the build.
  5. Install an operating system.
  6. Mount the server in a cabinet and cable it to a switch (see network).

Until all four core parts are in and the power supply can carry the load, the server will not boot. The rack picker warns you about missing parts before you mount it.

In the Tenant era, vendor orders arrive instantly and hands-on tasks take a few seconds at most. From the Established era on, tasks take real seconds: 30 for a processor, 12 for a memory module, 15 for a disk, 20 for a network card and 10 for a power supply. Vendor deliveries take 2 hours for most products and 24 or 48 hours for a few large ones. Premium halves every task, in the Tenant era too.

Chassis sizes

Current-generation chassis come in 1U, 2U and 4U heights. They differ in sockets, memory slots, drive bays, M.2 slots, PCIe slots and power supply bays.

Chassis Sockets Memory slots Drive bays M.2 PSU bays Price
1U Micro 1 2 2x 2.5" 0 1 $650
1U Single-Socket 1 4 4x 2.5" 1 2 $1,300
1U Storage 1 4 4x 3.5" 0 2 $1,500
2U Dual-Socket 2 8 8x 2.5" 2 2 $3,800
2U Storage (12 bays) 1 4 12x 3.5" 0 2 $2,600
4U Dense Compute 2 16 24x 2.5" 4 2 $6,500

The 1U models have two 1 GbE ports on board; the 2U and 4U models have two 10 GbE ports. A 3.5" bay also takes a 2.5" drive.

Older-generation chassis are only found second-hand:

Chassis Generation Socket Memory Drive bays M.2 PSU bays List price
1U Single-Socket 2 S2 4 × DDR4 4x 2.5" 1 2 $1,100
2U Dual-Socket 2 S2 8 × DDR4 8x 2.5" 0 2 $2,800
1U Single-Socket 1 S1 4 × DDR3 2x 3.5" 0 1 $800
2U Dual-Socket 1 S1 8 × DDR3 8x 2.5" 0 2 $2,000

All four have two 1 GbE ports on board. Second-hand prices sit well below these list prices (see market).

Processors and memory

Processors are described by cores, threads, base clock and TDP. Current models run from 4 cores and 8 threads ($350) to 64 cores and 128 threads ($7,500). Second-generation processors (socket S2) run from 6 to 28 cores, and first-generation ones (socket S1) from 4 cores and 4 threads to 12 cores and 24 threads; both are sold second-hand only. Threads matter because VPS capacity is sold in vCPUs, and a hypervisor sells two vCPUs per thread. A dual-socket board needs the same processor model in both sockets.

Memory must match the board's type (DDR3, DDR4 or DDR5) and run at a speed the board lists, such as 1333 or 1600 MT/s for DDR3 or 4800, 5600 or 6400 MT/s for DDR5. Every module in one server must share one speed: the game refuses a mixed set.

Faster memory pays off. A server's memory speed score compares its modules with the base speed of their type: DDR3-1333, DDR4-2666 and DDR5-4800 score 100, DDR5-5600 scores 117 and DDR5-6400 133. The score works three ways:

  • Orders: when a buyer chooses between your offers, each offer's weight is multiplied by the speed factor squared, so a 133 server is picked about 1.77 times as often as an otherwise equal 100 one.
  • Price: buyers accept half the extra speed as extra price. At 133 the market reference for that product is 16.5% higher before anyone walks away.
  • Happiness: customers gain 20 happiness points per 100% above base speed, about 7 at 133. Game communities feel it 1.5 times as much and software startups 1.2 times.

The score shows in the Resource use panel on the server's page.

Disks and RAID

Three kinds of drive exist: SAS disks (spinning, 7200 rpm) in 3.5" form (1 TB to 20 TB), SAS SSDs in 2.5" form (120 GB to 7.68 TB) and NVMe drives on M.2 (480 GB to 7.68 TB, including a 2 TB model). SAS disks and SAS SSDs sit in hot-swap caddies.

Every drive also brings I/O capacity, counted in customer units: 40 for an NVMe drive, 16 for a SAS SSD and 3 for a SAS disk. A group's capacity is the sum of its disks times a RAID factor: 1.0 for JBOD and RAID 0, 0.8 for RAID 10 and 0.6 for RAID 1 and RAID 5. Each VPS customer uses 1 unit and each hosting account 0.15; a dedicated tenant has the disks to itself. Two SAS SSDs in RAID 1 therefore carry about 19 VPS customers before sites slow down. Above 80% of capacity customers lose a little happiness; above 100% they lose much more (see selling hosting).

Identical disks (same type, form and capacity) form a group, and each group gets its own level:

Level Minimum disks Usable space Disk failures survived
JBOD 1 all disks 0
RAID 0 1 all disks 0
RAID 1 2, even count one disk all but one
RAID 5 3 all but one disk 1
RAID 10 4, even count half the disks half the disks

Changing the level wipes the installed operating system. To grow a RAID 1, 5 or 10 array without wiping it, swap its disks one at a time for larger ones of the same type and form from the server's Storage layout: each swap rebuilds the array onto the new disk while customers stay online (one redundancy fewer until it finishes), the array runs at its smallest disk's size until every disk is replaced, and keeps its level. When a disk dies inside the array's tolerance, the server keeps running as DEGRADED. When failures exceed it, the array is lost: virtual machines are destroyed, the OS is wiped and customers lose their data, unless the server is backed up.

Power supplies and the power budget

Power supplies run from 400 W to 2000 W in Bronze, Gold, Platinum and Titanium tiers, at 85%, 90%, 94% and 96% efficiency. The server's draw is a 30 W chassis overhead plus the chassis and every part inside it. Wall power is that draw divided by the supplies' efficiency, so a better tier lowers your power bill.

Supplies share the load, so their capacities add up. A server is N+1 redundant only when it still has enough capacity with its largest supply removed: two 750 W units carry 1,500 W, but the server is redundant only if it draws 750 W or less. Redundancy matters when a power feed drops, which is covered in power and cooling. Power supplies are hot-swap.

Operating systems and licenses

Software License Disk Notes
Server Linux LTS free 10 GB runs on every generation
Container Host OS free 12 GB generation 2 or newer
Hypervisor OS $15 per socket per month 20 GB runs virtual machines; generation 2 or newer
Hypervisor OS (legacy branch) $9 per socket per month 16 GB runs virtual machines on generation 1
Commercial Server OS $2 per core per month 32 GB generation 3 only
Storage Server OS $25 per server per month 16 GB turns the machine into a backup target

A paid license charges its first month at install and renews every billing month, which passes every 24 real hours; an unpaid renewal takes the server offline until paid. The hypervisor keeps 2 GB of memory for itself. The OS decides what the server can sell; see selling hosting.

Blade enclosures and nodes

A blade enclosure racks once and holds nodes: 6 nodes in 4U ($6,000), 8 nodes in 5U ($9,000) or 10 nodes in 5U ($12,000). The enclosure carries the power supplies (2 or 4 bays) and the network ports; nodes have no power supplies of their own and draw through it. Each node is built like a small server with its own processor, memory and up to two 2.5" disks plus one M.2 slot. Single-socket nodes cost $1,600 and dual-socket nodes $2,800.

Nodes cannot be racked alone and cannot run a hypervisor. Each one is sold as a dedicated server, and the enclosure only joins dedicated service groups.

Storage and backup servers

A storage chassis full of disks running the Storage Server OS becomes a backup target. It cannot be sold to customers. A storage chassis without an OS yet shows N/A for its own backup.

The quicker option is a rack NAS ($650): a closed 1U box with 500 GB of backup space on four drives in RAID 5, no parts to fit and no OS to install. Mount it, cable it to a switch and point servers at it like at a storage server. You point other servers at it, and it must hold the data in use on each of them: the OS, customer VMs, hosting accounts, and half the disk of a dedicated tenant. Servers are covered in the order you attached them until the space runs out. A protected server that loses its array keeps its customers' data; an unprotected one is wiped, and if it had customers you lose 5 reputation points. Storage servers and NAS boxes do not count towards your uptime, since they hold no customers.

Moving virtual machines

To empty a VPS server, use Move all virtual machines on its page. Every VM moves to another running hypervisor in the same datacenter with room for all of them, and the old server stops taking new customers. With a working backup the move is safe. Without one there is a 60% chance that one or a few customers, up to a third of those on the server, lose their data and leave; the game warns you and asks you to accept the risk first.

Resource use and limits

A racked server's page has a Resource use panel: processor, memory, disk, disk I/O, IPv4, bandwidth and power, each used against what the machine has, plus its customers, their monthly revenue, the load, the memory speed score and the customers' average mood. A server that sells to customers is at its limit, shown in amber on the Nodes page with one notification, once any of these is reached:

Resource At its limit when
Processor 95% of vCPUs sold, or of threads in use
Memory 95% used
Disk 95% used
Disk I/O customer load reaches what the disks can serve
Bandwidth average traffic reaches 90% of the port speed
Power peak draw reaches 90% of what its power supplies deliver
IPv4 no free address left where the server stands

Devices can also carry a label of up to 20 characters, set on the Nodes list and shown there, on the server's page and on the rack, in both the rack view and the floor view.

Wear and maintenance windows

Racked hardware ages 48 wear hours for every real hour. Health falls in a straight line over each part's lifespan: 35,000 hours for a SAS disk, 45,000 for a SAS SSD, 50,000 for NVMe, 60,000 for a power supply and 90,000 for processors, memory and network cards. A SAS disk therefore lasts about 30 real days. Busy servers wear up to 60% faster at full load, and hot cabinets raise failure odds further.

Disks fail most often, then power supplies, memory, network cards and processors. Below 40% health a part raises a maintenance alert and its failure odds climb steeply. The dashboard's resource forecast shows when each part will reach that point.

Drive bays, power supplies and blade bays are hot-swap. On a running server with customers, a power supply comes out only if the remaining supplies carry the load (once the server is shut down, any supply comes out), and a disk only if its array has redundancy to spare and the remaining disks still hold the data. Processors, memory, M.2 drives and PCIe cards are cold-swap: the game refuses to fit or pull them while the server runs, and on a server with customers what remains must still fit their virtual machines. A memory module can also be swapped for a larger spare from the inventory in one step, customers or not, once the server is powered down; it takes as long as pulling one module and seating another. The record of a stolen part can be cleared at any time, since nothing is physically there. To power a server down, schedule a maintenance window of 15, 30, 60 or 120 minutes at least 24 hours ahead. Scheduled maintenance does not count as downtime. A forced shutdown without a window does count, and the server powers itself back on after 60 minutes. A technician can replace failed parts for you.