IT & Managed Services Glossary
Wondering what SLA, MTTR or RTO means? This glossary gathers the definitions of essential technical terms used in managed services and IT operations.
Anti-affinity
Definition: A rule forbidding several virtual machines from running on the same physical host. It is essential as soon as two VMs form a cluster: without it, nothing stops the hypervisor from placing them side by side.
The trap: when both members end up on the same machine, redundancy becomes fictional — but the service answers, monitoring is green, and nothing raises an alert. The rule is often worked around during a maintenance window and never restored. That is why actual placement has to be verified, not merely the existence of the rule.
BCP (Business Continuity Plan)
Definition: An organisational and technical framework designed to maintain business operations with no or minimal interruption during a major incident. A BCP is more demanding and costly than a DRP, as it aims for zero perceptible downtime.
Objective: Immediate continuity with no perceptible outage. Typically requires active-active redundant infrastructure.
Also known as: PCA (Plan de Continuité d'Activité) in French-speaking contexts.
Guide: DRP vs BCP - which solution to choose? | → DRP/BCP Proxmox: recovery and continuity
DRP (Disaster Recovery Plan)
Definition: A set of documented procedures enabling an organisation to resume IT operations after a major disaster (fire, flood, cyberattack, critical hardware failure). Unlike a BCP, a DRP accepts a temporary interruption of service.
Also known as: PRA (Plan de Reprise d'Activité) in French-speaking contexts.
Guide: DRP vs BCP - which solution to choose? | → DRP Proxmox: multi-site architecture (RTO/RPO)
Failover
Definition: The automatic transfer of a service from one node to another when the first stops answering, with no human involvement.
Not to be confused with a restart: a failover moves the service to an instance that is already running — a few seconds, sessions sometimes preserved. A restart boots the machine elsewhere: a few minutes, and every in-flight session is lost. Both mechanisms exist; they do not answer the same need.
The 4 high-availability architectures | Keepalived or Pacemaker
Fencing (STONITH)
Definition: A mechanism guaranteeing that a node deemed faulty is no longer working, by cutting it off from the network, the storage or its power supply. The acronym STONITH stands for Shoot The Other Node In The Head.
Why it is indispensable: before restarting a service elsewhere, you need certainty that it is no longer running where it was — not a strong presumption. Without it, a merely unreachable node keeps writing while its replacement writes too. Many clusters ship with this function disabled because it complicates testing: a cluster without fencing is not a cluster, it is a gamble.
Managed Services
Definition: The full or partial outsourcing of IT system management to a specialised provider. Managed services typically include maintenance, monitoring, security and technical support, delivered on a subscription basis with defined SLAs.
Example: An SME outsources the management of its 5 servers to an MSP for EUR 70/month/server (7am-10pm, 7 days a week) or EUR 150/month/server (24x7) instead of hiring a full-time system administrator.
Also known as: Infogérance in French-speaking contexts.
Monitoring
Definition: Continuous automated surveillance of IT systems to detect anomalies, measure performance and prevent failures. Monitoring tools generate alerts when predefined thresholds are exceeded.
Common metrics: CPU, RAM, disk space, bandwidth, response time, service availability, SSL certificates, error logs.
MSP (Managed Service Provider)
Definition: A company specialising in the delivery of outsourced IT services. An MSP continuously manages, maintains and secures its clients' IT infrastructure, typically for a monthly subscription fee covering monitoring, patching, backups and incident response.
Related terms: Managed services provider, outsourced IT, IT outsourcing partner.
MTTR / Resolution Time SLA (Time to Resolve)
Definition: The maximum guaranteed delay between the reporting of an incident and the restoration of the service to a functional state. The resolution time SLA is the most binding commitment as it implies the problem is actually fixed, not just acknowledged.
Note: Resolution time is always longer than response time and commands higher costs. A 1-hour resolution SLA is very aggressive and expensive.
Typical values: 4h (critical), 8h (urgent), 24h (standard), "best effort" (low priority).
Also known as: GTR (Garantie de Temps de Rétablissement) in French-speaking contexts. In ITIL, also referred to as MTTR (Mean Time to Resolve/Restore).
On-Call Support
Definition: A period during which an engineer remains reachable and available to respond to incidents, typically outside business hours (nights, weekends, public holidays). On-call engineers are alerted via paging systems such as PagerDuty.
Example: 24/7 on-call support means an engineer can be contacted and intervene at any time, including at 3 AM on a Sunday.
Also known as: Astreinte in French-speaking contexts.
How our on-call works | See: On-call regulations in France (fr) | On-call support pricing | → Outsource on-call 24/7: complete guide
Operational Maintenance (Ops Maintenance)
Definition: The set of preventive and corrective actions required to keep an IT system running smoothly: OS and software updates, security patches, monitoring, proactive replacement of aging components and capacity planning.
Example: Operational maintenance of a server includes monthly security patching, disk space monitoring and proactive replacement of aging hardware components.
Also known as: MCO (Maintien en Conditions Opérationnelles) in French-speaking contexts.
Quorum
Definition: The minimum number of members that must see each other for a cluster to be allowed to decide. With three nodes a majority always exists; with two, neither can tell a dead neighbour from an unplugged cable.
The distinction that matters: for a stateless service — proxy, DNS resolver, NTP — two nodes are perfectly fine, the worst case being an address held twice for a few seconds. As soon as data is involved you need a third point of view: a lightweight witness that only votes, a fencing mechanism, or a full third node.
Split-brain and its three exits | Do you need three hypervisors?
RPO (Recovery Point Objective)
Definition: The maximum acceptable amount of data loss, measured in time, that an organisation can tolerate after a disaster. The RPO determines how frequently backups must be performed.
Examples:
- RPO 0: no data loss acceptable (synchronous replication)
- RPO 1h: up to 1 hour of data loss is acceptable
- RPO 24h: daily backup is sufficient
RTO (Recovery Time Objective)
Definition: The maximum acceptable duration of downtime for a system or service after a disaster. The RTO determines the recovery architecture required.
Examples:
- RTO 0: no downtime (automatic failover)
- RTO 4h: recovery within 4 hours
- RTO 24h: recovery within 1 business day
Guide: DRP/BCP with RTO/RPO | Calculate the cost of downtime
SLA (Service Level Agreement)
Definition: A formal contract or contractual annex that defines the expected service levels between a provider and its customer. An SLA typically includes response times (TTR), resolution times (MTTR), uptime guarantees and penalties for non-compliance.
Key components of an SLA: scope, support hours, response/resolution times by priority level, guaranteed uptime (99.9% = 8h45min downtime/year), penalties, exclusions.
Complete guide: SLA, SLO and SLI | Guide: Response Time, Resolution Time and SLA
Split-brain
Definition: A situation where two cluster members simultaneously believe they are legitimate. It happens when the network separates them without either one failing: each observes the other's silence and concludes it must take over.
Its cost depends on what the cluster carries: stateless, both nodes hold the address for a few seconds — annoying, not destructive, and it resolves itself. With data, both write in parallel and the datasets silently diverge: repairing it is not a matter of restarting, but of choosing which version you throw away.
TAMS (Third-Party Application Maintenance Services)
Definition: The outsourcing of business application maintenance to a third-party provider. TAMS covers bug fixes, minor enhancements and user support for a specific application (ERP, CRM, custom software).
Difference from managed services: TAMS focuses on business applications (ERP, CRM, custom software), while managed services focus on infrastructure (servers, networks, operating systems). On a hosted ERP the two converge: see our Odoo managed services (server + PostgreSQL database).
Also known as: TMA (Tierce Maintenance Applicative) in French-speaking contexts. Also referred to as Application Management Services (AMS).
Ticket (Ticketing)
Definition: A support request or incident report logged in an IT Service Management (ITSM) system. Each ticket has a unique identifier, a status, a priority level and a complete history of interactions.
TTR / Response Time SLA (Time to Respond)
Definition: The maximum guaranteed delay between the reporting of an incident and the start of its handling by a qualified engineer. The response time marks the moment someone begins actively working on the problem.
Important: The response time SLA does not guarantee resolution of the problem, only that work has begun. It is the resolution time (MTTR) that guarantees service restoration.
Typical values: 15 min (critical), 1h (urgent), 4h (standard), 8h (low priority).
Also known as: GTI (Garantie de Temps d'Intervention) in French-speaking contexts. Also referred to as Time to Acknowledge (TTA) or First Response Time.
Complete guide: Response Time, Resolution Time and SLA | → Outsourced on-call: 4h vs 1h response in practice
VIP (virtual IP address)
Definition: An IP address that belongs to no machine outright: it is carried by whichever one is currently providing the service, and moves to another on failure. Also called a floating address.
At a hosting provider, its mobility depends on the provider: on a layer-2 segment where you control MAC address learning, VRRP moves it within one to three seconds. Otherwise it has to go through the provider's control plane, and failover is counted in tens of seconds.
Load balancing
Definition: Distributing requests for one service across several instances. Not to be confused with high availability: that answers “what happens when a component fails?”, load balancing answers “what happens when everything works, but there is too much of it?”.
What actually matters: The two overlap heavily — a shared address that fails over is one step from being an address that distributes — but a balancer checks nothing unless asked to, and above all capacity does not fail over on its own: a survivor inheriting everyone else's load falls in turn.
Anycast
Definition: One IP address advertised from several places at once: routing decides which instance answers, with no client-side configuration at all.
What actually matters: Advertisement scope is a parameter, not a property. Kept inside the private network it needs neither an AS number nor transit and runs on OSPF or IS-IS; propagated to the Internet it requires an AS, an advertisable prefix and transit providers. Either way the network checks nothing: an instance that advertises without answering becomes a silent black hole.
Round-robin DNS
Definition: Publishing several addresses behind a single name, so clients do not all head for the same machine.
What actually matters: DNS does not decide the distribution, it proposes an order. BIND rotates that order on every answer — since 9.21.14 random ordering was removed and random is an alias for cyclic. The resolver then caches the address set for the whole TTL, so the real granularity is “per resolver”, not “per client”. And the client does as it pleases: chrony in pool mode keeps several, in server mode only one.
VRRP
Definition: An IETF standard (RFC 5798) that lets a group of machines share one IP address: one advertises it, the others wait and take over if it goes quiet.
What actually matters: VRRP moves an address, it synchronises nothing. Application state, in-flight sessions and data remain entirely your responsibility. Another field trap: the virtual router identifier must be unique across the broadcast domain, not merely within your own configuration — two neighbouring clusters reusing the same number will elect against each other.
Layer 4 and layer 7
Definition: The level at which a balancer makes its decision. At layer 4 it sees a connection — an address, a port — and never opens it. At layer 7 it reads the request and can choose on the URL, a header or a cookie.
What actually matters: This is an axis distinct from the depth of the health check, and the two are often confused. Keepalived can check a target over HTTP, HTTPS or by script while distribution stays at layer 4, inside the kernel. So you do not move to layer 7 to get an application-level check: you move there to decide per request, which layer 4 cannot do since it never opens the connection.
Health check
Definition: A repeated probe that removes a target from rotation when it stops answering, and puts it back when it returns.
What actually matters: Its worth depends entirely on what it verifies. A TCP check is passed without difficulty by a server that accepts the connection and then returns an error on every request. And a /health endpoint that returns 200 without checking anything is no better than a port test — you paid for layer 7 and got a layer 4 check. At the other end, DNS and routing check nothing at all.
Sticky session
Definition: A mechanism that steers a given client to the same instance, usually through a cookie set by the balancer.
What actually matters: It is not a guarantee: it holds as long as the cookie is valid and the instance is available — if that instance goes down, the session is lost anyway. It is a patch over an application keeping state in local memory: distribution becomes uneven and the failure of one instance disconnects every user attached to it. Where possible, move the state out of the application.
Local loop
Definition: The network segment between the customer premises and the operator's first concentration point. On copper, it is the twisted pair running from the wall socket to the local exchange. On fibre, it is the optical local loop between the optical termination outlet and the optical distribution node.
What actually matters: it is the least redundant segment of the whole chain, and the one that accounts for most outages — roadworks, a digger, a building incident. Operator core networks are meshed and reroute on their own; the local loop is a single strand. Two subscriptions with two different operators very often ride the same local loop: competition happens on the service, not on the civil engineering.
Mutualisation point (PM)
Definition: In the French FTTH regulatory model, the point where the infrastructure operator that deployed the shared optical local loop opens access to commercial operators. Each commercial operator connects its own equipment there, then serves its customers over the shared fibre downstream. Depending on how dense the area is, the PM sits inside the building or in a street cabinet.
What actually matters: it is a mandatory, non-redundant transit point. Two FTTH links bought from two different operators at the same address almost always pass through the same PM, and often through the same building entry duct. The customer believes they have doubled their access; they have doubled the invoice and the contract, not the physical path. This is why a connectivity audit asks for the PM, not just the operator's name.
FTTH (Fiber To The Home)
Definition: Fibre to the home, over a shared optical local loop. The architecture is usually a tree (PON): one optical tree serves several dozen subscribers who share the capacity. This is the consumer offering.
What actually matters: the advertised speed is a maximum, not a commitment — it is shared and not guaranteed, and upstream is usually far below downstream. More importantly, a consumer FTTH offer carries no operator restoration commitment: after an outage, the line is repaired on a best-effort basis with no contractual deadline. Acceptable as a backup link; never as the only access of a site hosting production.
When FTTH is enough. If your production does not run on your premises but on virtual machines hosted on our Nimbus platform or on a managed Proxmox hypervisor, your customers never travel over your office access to reach the service: it is served from an Equinix datacenter in the Paris region, on our own AS206014 network. Your site link then only carries your team's access — FTTH with a 4G/5G backup often covers that. The full reasoning sits in the FTTO entry.
FTTO (Fiber To The Office)
Definition: Dedicated point-to-point fibre to the company site: a fibre that belongs to you alone between the site and the operator's point of presence, with no shared capacity. Symmetrical guaranteed bandwidth, contractual operator restoration time, static IP addresses, operator-side monitoring.
What actually matters: the price gap with FTTH — often tenfold — does not buy bandwidth. It buys the bandwidth guarantee, the restoration commitment, and the fact that nobody else shares the infrastructure. Comparing FTTO and FTTH on the headline speed alone is meaningless: they are two different products that happen to carry bits.
Where does your production actually run? That question settles FTTH versus FTTO, not the bandwidth figure. If your servers sit on your premises, the site link is the production link: it carries your customers, your backups and your on-call alerting, and it justifies FTTO with its restoration commitment. If your production runs on virtual machines hosted on our Nimbus platform or on a managed Proxmox hypervisor, the link carrying the service is ours — in an Equinix datacenter in the Paris region, on our own AS206014 network. Your customers never travel over your office access to reach the service.
Your site link then only carries your own team's access. An outage becomes inconvenient rather than blocking for the outside world: FTTH with a 4G/5G backup often covers the need, and the budget moves from access to architecture — where it buys real availability instead of a penalty clause.
FTTE (Fiber To The Enterprise)
Definition: A business fibre offering positioned between consumer FTTH and dedicated FTTO. Depending on the operator, it means either an offer built on the shared FTTH infrastructure but sold with business commitments — symmetrical bandwidth, a restoration time, a named contact — or an entry-level dedicated fibre.
What actually matters: the acronym has no single definition on the French market. Two “FTTE” offers from two operators may not be the same product at all. Three questions settle it, and none of them is about the acronym: is the link shared or dedicated? Is the bandwidth guaranteed or best effort? Is the restoration time contractual, over which hours, with what penalties? Get the answers in writing before signing.
Operator GTR (carrier restoration time)
Definition: A guaranteed restoration time bought from a telecom operator on a link: the maximum contractual delay between reporting an outage and the link being restored. Common values: 4h during business hours, 4h 24/7, 10h, or next business day. That single line, far more than the bandwidth, explains most of the price gap between two links.
What actually matters: an operator restoration commitment covers the link, not your service. It says nothing about your servers, your applications or your data. Two details to read in the contract: the clock starts when the incident is reported, not when the link went down — a link that failed at 2am and was reported at 9am burns a window starting at 9am; and the penalty is almost always capped at a fraction of the monthly fee. A restoration commitment compensates, it does not restore.
Why an operator commits to a restoration time and we do not. On a link, the failure modes are bounded and known: a cut fibre, an optical terminal to swap, a faulty line card. Repair time is statistically predictable, therefore insurable. A system incident has no such property — until the diagnosis is made, the same apparent failure may take two minutes or turn out to be data corruption. That is why we commit to a response time (GTI) on operations, and not to a restoration time.
Path diversity
Definition: The property of an architecture where two links share no physical element: no entry duct, no cable chamber, no mutualisation point, no distribution node, no termination equipment. It is the only form of access redundancy that survives a digger.
What actually matters: two operators do not make two paths. This is the most expensive confusion in the field, because it only surfaces on the day of the outage — both links drop together and nobody understands why. What to ask before signing: the route, the distribution node, the mutualisation point, and the building entry point. Real diversity usually requires two physically separate building entries, which is civil engineering measured in weeks, not a checkbox.
The fallback: when a second building entry is not feasible, a fibre link paired with a radio link — 4G/5G or microwave — gives media diversity in place of route diversity. The backup bandwidth is lower, but the two links cannot be severed by the same shovel.
→ Multi-site and internal anycast | High-availability architectures
Need expert guidance?
RDEM Systems helps you set up your managed services with SLAs tailored to your needs. Contact us for a free assessment.
In-depth Guides
Response Time, Resolution Time & SLA: complete guide
Understand and negotiate your service level commitments
DRP and BCP: which solution?
Choosing between disaster recovery and business continuity
Server management: complete guide
Everything you need to know about outsourced server management
MSP: definition and complete guide
Understanding the Managed Service Provider model