{"canonical_url":"https://engineer.company/glossary/cloud/","content_markdown":"# Cloud\n\n## What it is\n\nRenting computers you will never see is the shortest honest description of the\ncloud, known in full as cloud computing: a provider owns the buildings, the\npower and the hardware, and the customer switches servers, storage and networks\non and off through a web console or an API, paying for what was used. Every\n“cloud” workload still runs on a physical machine in a data centre; someone else\nbought it and rents slices of it to many customers at once.\n\nA precise definition comes from the US National Institute of Standards and\nTechnology. Its Special Publication 800-145, written by Peter Mell and Timothy\nGrance and published in September 2011, describes cloud computing as a model for\n“on-demand network access to a shared pool of configurable computing resources”\nthat can be provisioned and released with minimal management effort. It names\nfive characteristics. On-demand self-service means a new server without talking\nto anybody. Broad network access means the service is reached over the network\nfrom ordinary devices. Resource pooling means customers share the provider's\nhardware and generally choose a country or a data centre, never a rack. Rapid\nelasticity means capacity grows and shrinks with demand. Measured service means\nusage is metered, which is what makes a bill by the hour or the gigabyte\npossible.\n\nA dedicated server rented by the year, changed by ticket and billed at a flat\nrate meets few of these tests. That is hosting, often a good choice, but a\ndifferent contract with different economics.\n\nThe same document sorts cloud services into three models. With infrastructure as\na service (IaaS) the customer rents virtual machines, disks and networks, and\npatches the operating system on them. With platform as a service (PaaS) the\ncustomer deploys code onto the provider's runtime. With software as a service\n(SaaS) the customer uses a finished application, such as web mail. It adds four\ndeployment models: public, open to anyone; private, provisioned for one\norganisation and possibly on its own premises; community, shared by\norganisations with common concerns; and hybrid, two or more of these bound\ntogether so data and applications can move between them. The best-known public\nclouds are large platforms such as AWS, Azure and Google Cloud, and beside them\nsmaller providers sell plain virtual machines at simpler prices.\n\n## Why it matters\n\nThe cloud turns a purchase into a running cost, and that is most of the decision\nin one line. Buying servers means paying up front for capacity sized to a guess\nabout next year's busiest day. Renting means paying every month for what is\nswitched on, with the right to change the answer next week. For a new product or\na seasonal business, that flexibility is worth a premium. For a workload that\nhas run at the same size for three years, the premium keeps being charged long\nafter anyone used the flexibility.\n\nThe second decision is how far up the stack to rent. A virtual machine is close\nto a server a team already understands, and it moves between providers with\nlittle more than a copy of the disk. A managed database, a serverless function\nor a proprietary data warehouse takes work off the team, and each one is also a\npiece of the application written against one company's interface. The useful\nhabit is to know what leaving each managed service would cost, and to accept\nthat cost on purpose rather than discover it during a price rise.\n\nThe third is responsibility. Renting hardware hands over the building, the power\nand the hypervisor; who can log in, which ports are open and whether the data\nhas a copy elsewhere stay with the customer. The large providers each publish a\nshared responsibility model that draws this line, and the accounts, the\nconfiguration and the data sit on the customer's side of every version.\n\n## What it looks like in practice\n\nFor one international client the subject was the bill.\n[Their AWS setup had been over-provisioned and had collected services they no longer used](/portfolio/a-tenth-of-the-cloud-bill-and-nothing-lost/),\nso the monthly cost had drifted far out of proportion to the business. The work\nbegan with an audit of what was really used, removed unnecessary services,\nadjusted the remaining capacity and moved workloads away from AWS. Those are\nthe infrastructure changes described by the engagement record; it does not\nestablish a like-for-like cost ratio, a productivity result or a disruption-free\nmigration.\n\nThe company's own public-services host sits at the other end of the scale.\n[It is a single-core cloud instance with 512 MB of memory and 10 GB of disk](/portfolio/a-whole-company-on-one-small-server/),\nabout 464 MB of it usable, serving TLS for seven domains, a git forge, a Tor\nonion service, Gemini and Gopher servers and intrusion banning. Backup\nautomation is implemented separately; its configuration does not establish\nactive encrypted backups. In a cloud console the easy answer to a list that\nlong is a bigger instance type, and the onion service went on without one:\nover a recorded day of samples, the instance's lowest reading of available\nmemory was about 310 MB of 464, its swap use 2.6 per cent and its single core\n99.7 per cent idle. The machine costs less per month than a lunch, and mail is\ndeliberately kept off it because a mail server needs headroom this one has\nalready spent.\n\nRenting also means deciding how recovery extends beyond the provider's copy. The company's\n[backup and restore-test implementation](/portfolio/backups-that-have-actually-been-restored/)\nsupports a database dump where configured, then an encrypted, deduplicated restic snapshot\nover SFTP to storage at another provider, retention pruning and an integrity check. Backup\nand restore have separate success signals. A separate monthly workflow brings a known file\nback to a scratch location and compares its bytes, testing the return trip rather than just\nthe upload. Timers depend on a configured repository: the implementation record dated\n2026-10-27 says they were disabled without one and records no completed live restore.\nThat is an implemented recovery design with activation prerequisites, not proof of a\nsuccessful recovery or of the current runtime state.\n\n## Where it goes wrong\n\nClouds fail less often than the habits around them, and the common problems are\nslow, quiet and expensive.\n\n**The bill grows while the product stays the same size.** Resources accrete: a\ntest server left running after the test, a disk still billed after its machine\nwas deleted, snapshots kept forever, a load balancer in front of nothing. Moving\ndata out of a provider, or between its regions, is metered too, so shipping logs\nand backups across those boundaries pays on every copy. Tag every resource with\nan owner and a purpose, set a budget alert, and read the bill line by line once\na month.\n\n**Capacity is sized for a day that never comes.** Machine sizes are picked once,\nat launch, from a guess about the worst case, and the guess outlives the people\nwho made it. Measure what a workload uses across a normal week and a busy one,\nsize to that with a margin, and let the autoscaler shrink as readily as it\ngrows.\n\n**The provider's snapshot is treated as the backup.** It lives in the same\naccount, behind the same credentials, at the same company as the machine it\nprotects, so a stolen login or a suspended account takes the server and its only\ncopy together. A real backup sits under different credentials, ideally at a\ndifferent provider, encrypted with a key that provider never holds, and is\nrestored on a schedule.\n\n**Security stays where the defaults left it.** The usual openings are a firewall\nrule admitting SSH from the whole internet, a long-lived access key in a\ndeveloper's dotfiles, an administrator account used for everyday work and a\nstorage bucket made public for one demo. Grant the least access that does the\njob, prefer short-lived credentials, require a second factor on every human\nlogin, and keep the configuration in code so that a change to it gets reviewed.\n\n**Leaving turns out to need a rewrite.** Every proprietary queue, function\nruntime and database dialect ties code to one provider a little more firmly.\nKeep the core on open interfaces where that is cheap, such as PostgreSQL,\ncontainers and plain virtual machines, and write down the exit cost of each\nproprietary service when it is adopted.\n\n## Further reading\n\n- [The NIST Definition of Cloud Computing](https://csrc.nist.gov/pubs/sp/800/145/final)\n  — the short definition with its five characteristics, three service models and\n  four deployment models.\n- [Guidelines on Security and Privacy in Public Cloud Computing](https://csrc.nist.gov/pubs/sp/800/144/final)\n  — what an organisation should weigh before moving data, applications and\n  infrastructure to a public provider.\n- [NIST Cloud Computing Reference Architecture](https://www.nist.gov/publications/nist-cloud-computing-reference-architecture)\n  — the architecture and taxonomy NIST built on its definition, naming the\n  parties to a cloud arrangement.\n- [FinOps Framework](https://www.finops.org/framework/) — the FinOps\n  Foundation's model for making cloud spending a shared practice of engineering\n  and finance.\n\n\u003chttps://engineer.company/glossary/cloud/\u003e\n","date_modified":"2026-11-08T02:22:23Z","description":"What the cloud means in NIST's own terms, when renting beats owning, and the bills, defaults and backups that catch out the teams who rent it.","formats":{"gemtext":"https://engineer.company/glossary/cloud/index.gmi","gopher":"https://engineer.company/glossary/cloud/gophermap","html":"https://engineer.company/glossary/cloud/","json":"https://engineer.company/glossary/cloud/index.json","markdown":"https://engineer.company/glossary/cloud/index.md","pdf":"https://engineer.company/glossary/cloud/index.pdf"},"kind":"page","language":"en","title":"What is the cloud?","url":"https://engineer.company/glossary/cloud/","version":1}