Dedicated hosts running no foreign guests, with a hypervisor operated by ENTRONYX CLOUD: 5 host classes under 4 stacks at 4 locations in Germany. This page also tells you when this approach is the wrong one.
Nobody looks for a Private Cloud just because it is one. You look for it because an invoice has gone up, a room has run out of space, or an auditor asks a question. For three of these scenarios, there is a specific cluster here — with the amount the configurator calculates for it, and the link that opens it exactly like that.
Initial situation
Since the switch to per-core subscriptions, you pay a multiple for the same hosts, the minimum purchase per socket hits small clusters in particular, and changing the hypervisor during operations is not an option.
You keep vSphere, vCenter and your runbooks. The licence runs on our contract and is shown transparently on every invoice — per core, minimum 16 per socket, which does not apply with 48 cores per host, and with no minimum term. We patch the hypervisor, during the maintenance window, host by host.
Monthly net
€5,199.11$6,030.97
If you want to move away from the VMware licence, you get the same hosts under the open KVM stack — €49.00$56.84 per socket instead of €9.90$11.48 per core.
Initial situation
The hardware is depreciated, the maintenance contracts are expiring, the UPS is older than the applications it protects. Building a new facility is not worth it for two racks — but the data should stay in Germany, with an operator you can reach.
A hyperconverged cluster replaces servers, storage system and storage network in one step: three nodes, a distributed NVMe pool, growth by individual nodes. We maintain firmware, hypervisor and fabric; your administrators keep the guests.
Monthly net
€8,632.88$10,014.14
Under 6 nodes, the pool runs doubly mirrored — 47% of the raw capacity is then usable instead of 62%. The nightly backup goes to a second EU region; if the copy is also to be located in Germany, the “Asynchronous Replication” level is the right one — adjustable in the configurator.
Initial situation
The base load runs around the clock, but is billed by the hour like a peak. The next audit wants to know which data is on which device and who has access — an answer that a shared platform does not provide at the device level.
Managed dedicated hosts under the open hypervisor: named devices in a named rack, management via API and Terraform, billing per host and month. Fluctuating loads stay in the Public Cloud — both are connected to the same vRack.
Monthly net
€2,438.91$2,829.14
A single host is enough to start; an automatic restart after a host failure is only available from the second one onwards.
All amounts net per month, location Frankfurt, before term discount — 8% for twelve, 14% for twenty-four months. The configurator calculates with the same rates and shows each item individually; location, storage and operations level can be changed there without starting from scratch.
Choose all five steps yourselfYour scenario doesn't fit? We will call you back — within 4 hours during business hours (voluntary commitment, not part of the SLA).
Three steps lie between the first conversation and the first running machine. The first is the hardest, because it can also turn out against the Private Cloud. Next to it is what the step is about: the one block that belongs to you, and the line where it ends; an open rack with the compute nodes above the storage; a second, identically built block, connected to the first.
The Private Cloud is an operating model, not a scale. Across 12 layers, it sits between the Public Cloud and bare metal: more control than in the Public Cloud, less in-house operation than on bare metal.
The first step is therefore not a cart, but a demarcation. Where the load fluctuates daily or no hypervisor is needed at all, a different line is right — the cross-check on this page calls these cases by name instead of paraphrasing them.
5 host classes from 16 to 384 cores carry 4 stacks. The stack determines the licensing model and the smallest permissible cluster: ENTRONYX Managed vSphere requires 2 hosts, ENTRONYX HCI 3.
Storage is the second decision and the more consequential one. A datastore is the storage area where the hard drives of the virtual machines reside. Each datastore class carries a commitment for latency and IOPS (input and output operations per second) — and the condition under which the commitment applies.
Platform operations are included in the host price: firmware, hypervisor, fabric — the network between hosts and storage — and storage controllers are on us. Above this are 2 further operating levels that push the boundary upwards layer by layer.
Recovery is decided separately. 4 levels range from nightly backup to a second region to synchronous mirroring between two locations; each specifies a recovery point and recovery time as a commitment.
The summary of the third step: what each tier promises, where it replicates to, what it costs per month and how the promise is audited. The conditions under which a number applies are on the recovery page — where they are part of the promise.
| Tier | Recovery point | Recovery time | Target region | Monthly amount | Verification |
|---|---|---|---|---|---|
| Secured recoveryKVM Platform, Managed vSphere, HCI, SAP Certified | 24 h | 8 h | Second EU region, from 300 kmfrom fra1: Munich (muc1) | €149.00$172.84per environment, plus €0.0074$0.0086 per GB | 1 Recovery test per yearwith report to you |
| Pilot LightKVM Platform, Managed vSphere, HCI | 4 h | 4 h | Region in the same jurisdiction, from 200 kmfrom fra1: Munich (muc1) | €490.00$568.40per environment, plus €0.0190$0.0220 per GB | 2 Recovery tests per yearwith report to you |
| Asynchronous replicationKVM Platform, Managed vSphere, HCI, SAP Certified | 15 min | 2 h | Region in the same jurisdiction, from 200 kmfrom fra1: Munich (muc1) | €990.00$1,148.40per environment, plus €0.0340$0.0394 per GB | 4 Recovery tests per yearwith report to you |
| Synchronous metro mirroringManaged vSphere, SAP Certified | 0 minsynchronously mirrored | 20 min | Metro pair in the same metropolitan area, 5 to 60 kmfrom fra1: Frankfurt am Main (fra2) | €1,890.00$2,192.40per environment, plus €0.0690$0.0800 per GB | 4 Recovery tests per yearwith report to you |
Net amounts per month: a base fee per protected environment and a rate per protected gigabyte. A tier does not replace an availability promise — and the availability promise does not replace a tier.
Conditions, target locations and testing proceduresFive categories: the platform itself, certified environments for applications with their own requirements, storage and backup, network separation and operations on top.
Three ways to your own virtualisation on hardware that no one else uses — managed, hyperconverged or as a pure host.
from €461.59$535.4413 quotesFor applications that require written evidence of their hardware — above all SAP HANA.
from €2,313.00$2,683.089 quotesHow fast the storage responds, how often it is backed up and where the second copy is located in an emergency.
12 quotesHow your hosts are connected to each other, to other locations and to the Public Cloud — and where responsibility changes.
5 quotesWho patches, when maintenance happens and who you reach in an emergency — all in the contract, not just goodwill.
8 quotesConfigurator for this section
The three factors that drive the cost. The summary calculates with the same rates as the configurator and adds storage, network and operations by default; ENTRONYX Managed vSphere starts at 2 × PC-M64, the stack table below lists €2,996.40$3,475.82 for this without storage. The link in the summary opens the configurator with exactly this choice.
2–32 nodes under ENTRONYX Managed vSphere
Balance
vSphere cluster · 2 × PC-M64
96 cores · 464 GB allocatable · Frankfurt am Main
vSphere cluster · 2 × PC-M64: €3,291.31 net per month
Included: vRack 1 Gbit/s · 1 Gbit/s guaranteed · 1 × IPv4 + /64 IPv6 · Basic DDoS protection · Platform operations · Unlimited traffic
Monthly net
€3,291.31$3,817.92
Frankfurt · before term discount · storage, network and operations on default
Public Cloud, Private Cloud and Bare Metal run in the same halls on the same device family. The difference is not the location or the form factor, but who a device is assigned to, how it is billed and from which layer you are responsible.
In the Public Cloud, capacity comes from a pool. A host carries guests of multiple customers. They are separated by the hypervisor, the software that divides a device into virtual machines. We operate and allocate the cluster behind it. Billing is based on term — from €0.0082$0.0095 per hour. If you switch off at night, you don't pay at night.
In the Private Cloud, the host is a named device. Only your guests run on it; placement is not decided by us based on utilisation, but by you according to your rules. Billing is per host and month, regardless of how many virtual machines are on it. We operate what lies below the guest — hypervisor, firmware, fabric.
On bare metal, our responsibility ends at the hardware. You get a device and the remote management card; you alone decide which operating system, which hypervisor and which firmware version run on it. This is the greatest freedom of the three cuts and at the same time the greatest operational burden.
What is the same in all three cuts
What only the Private Cloud provides
Instances on shared hosts, billed by the hour, created and removed in minutes.
Dedicated hosts with managed hypervisor. You allocate the capacity, we keep everything below it in operation.
The entry level applies to PC-S32 under ENTRONYX KVM Platform — a single host. ENTRONYX Managed vSphere starts at 2 hosts.
One device, one remote management card, nothing else. Everything above the hardware belongs to you — including the operational burden.
The line on which the decision usually turns is not the price, but the firmware: in the Public Cloud and in the Private Cloud it lies with us, on bare metal with you. Below that nothing changes, above that everything.
Data centre, power, cooling
Hardware and spare parts
Firmware and BIOS
Hypervisor and cluster services
Storage fabric and datastores
Network segmentation and firewall rules
Capacity planning and placement
Guest operating system
Middleware and application
Application data and classification
Access management and roles
Backup and recovery test
The same comparison, condensed to the rows that are actually asked about in a procurement discussion. The table recommends nothing — it places side by side what each model guarantees and what it costs.
| Feature | Public Cloudshared hosts | Private Clouddedicated hosts | Bare Metalbare metal |
|---|---|---|---|
| IsolationWho shares the metal under your machine? | Shared hosts. The separation is provided by the hypervisor; with dedicated flavours, the core is reserved for you, but the device is not. | Dedicated hosts under your ID, no foreign guests on the same metal, management plane in your own tenant. | Dedicated device without a virtualisation layer above it — neither foreign guests nor a hypervisor computing alongside. |
| Licence sovereigntyWhose contract does the software between metal and guest run on? | No own licence. What the platform needs is included in the instance price and is not visible from the outside. | On our contract, openly stated: €9.90$11.48 per core and month under vSphere, €13.90$16.12 under HCI. | Your licences on your contract — including the audit obligation towards the vendor and the question of how many cores an audit counts. |
| Operational effortHow many layers below your application remain with us? | 6 with us, 3 shared, 3 with you — out of 12 layers. | 5 with us, 5 shared, 2 with you — out of 12 layers. | 2 with us, 0 shared, 10 with you — out of 12 layers. |
| Price per core and monthCalculated on the machine under PC-M64 — the same hardware in each column, as far as the model sells it that way. | Cores are not the unit there. ND6 costs €16.69$19.36 per dedicated vCPU (virtual core) and month, hourly €0.0200$0.0232. | €31.21$36.20 per physical core, licence included — PC-M64 under vSphere. | €17.58$20.39 per physical core as hardware share of the same machine — without hypervisor, without licence, without operations above it. The catalogue does not currently list it as a separate Bare Metal model. |
| ElasticityIn what steps and at what pace does the environment grow? | Billed hourly, capacity from a pool. A powered-off instance costs no compute time. | Monthly price per host, expansion in host steps. A cluster starts at 2 hosts under vSphere and 3 under HCI — below that there is no automatic restart. | Monthly price per device, expansion in device steps. No cluster requirement, but also no instance that absorbs the failure of a device. |
Four workloads that typically run on managed hosts. Each row lists the series, the configuration with the amount the configurator calculates for it, and the rule by which it is sized. User numbers are not listed here: how many sessions or bookings a host supports depends on your application, not the hardware.
Database and application server as separate guests. The database depends on write latency and single-core clock speed: PC-M64i clocks up to 3.8 GHz, the Datastore NVMe Performance guarantees 0.4 ms p99 latency.
Sizing: Two hosts, and the entire environment fits on one of them: if one host fails, ENTRONYX Managed vSphere restarts the guests on the other. Sizing is therefore against one host — 232 GB memory and 288 vCPU allocatable —, not against the total.
SAP HANA does not belong in this cluster, but on certified nodes with their own protocol. SAP nodes
Monthly net
€3,402.64$3,947.06
High capacity with low compute load. The capacity datastore places an NVMe cache in front of HDD capacity and keeps snapshots for 30 days — sufficient for file services and quiet volumes.
Sizing: The decisive factor is the daily volume of data moved, not the total amount. The NVMe cache covers 8% of the datastore size. Working sets beyond this pass through to the HDD tier; the 4 ms latency guarantee then no longer applies.
Two hosts: if one fails, the file server restarts on the other — with one host, this restart would not happen.
Monthly net
€2,430.72$2,819.64
A target for your backup software, physically separated from the systems it backs up. Capacity is the main requirement; PC-M64 is the most affordable class that supports a shared datastore in Berlin.
Sizing: Berlin is around 420 km as the crow flies from Frankfurt am Main — further than the minimum distance of 300 km required by the "Secured recovery" level. Upper size limit: data volume times the number of full backups kept; the datastore grows up to 512 TB in the configurator.
One host is sufficient for a backup target; automatic restart after a host failure is only available from the second host onwards. If the Private Cloud itself is to be backed up, the “Secured recovery” tier is the shorter route. Recovery tiers
Monthly net
€1,309.57$1,519.10
Many concurrent sessions: here, the number of cores counts more than their clock speed, and maintenance must not interrupt logins. PC-L128 provides 64 cores in one socket; ENTRONYX Managed vSphere evacuates the hosts one after the other before maintenance.
Sizing: How many sessions a vCPU can handle depends on your application — which is why there is no user count here. The upper limits per host are fixed: 472 GB memory and 256 vCPU allocatable. With two hosts, the load of one must fit on the other, even during maintenance.
Monthly net
€5,138.23$5,960.35
Net amounts per month, before term discount, without disaster recovery and without operations level above platform operations. Each configuration can be further adjusted in the configurator — capacity, location and number of nodes, without starting from scratch.
Four cases where another part of our catalogue gives the better answer. They are listed here because an offering that only names its strengths becomes expensive in operation — and because otherwise the decision is only made in the third conversation.
A managed host is billed per month, not by runtime. If you size for the peak, you pay for the peak at night and on weekends too. There is no shutdown that lowers the host price — the hardware sits in the rack under your name, whether it computes or not.
The right choice here is the Public Cloud: it is billed hourly and capacity comes from a pool instead of a booked device.
Shared instance from €0.0082$0.0095 per hour versus a host that is billed monthly.
To Public CloudPlatform operations are fixed in the host price. If you do not use the hypervisor, you still pay for its operations, its maintenance window and — depending on the stack — its licence. For a single database on bare metal, this is a surcharge with no equivalent value.
The right choice here is a dedicated server: the same bare metal, without the virtualisation layer and without its operations surcharge.
FORGE-32 as a dedicated server: €220.16$255.39. The same machine as PC-S32: €461.59$535.44. The difference of €149.00$172.84 is the platform operations.
To Bare MetalIn the Private Cloud, BIOS, firmware, hypervisor, fabric switches and storage controllers are our responsibility. That is the core of the offer — and at the same time its limit: you cannot set your own hypervisor build, your own kernel below the guest or a self-chosen firmware version here.
If you need to fill these layers yourself, you need dedicated hardware without an operations contract on top.
4 layers are non-negotiably with us: Hypervisor, BIOS and firmware, fabric switches, storage controllers.
View shared responsibilityAn automatic restart after a host failure requires a second host and a shared datastore. If you can only pay for one host, you get no high availability — not even in a small version. For the same reason, the smallest class is explicitly not approved for shared datastores.
The right choice then is a single instance with a backup, not half a cluster with the promise of a whole one.
ENTRONYX Managed vSphere requires 2 hosts, ENTRONYX HCI 3. A dedicated server starts at €65.54$76.03.
To VPS0105Allocation
A managed host is a named device in a named rack. It runs exclusively your guests, and placement follows your rules, not our utilisation. That is why the price is tied to the host and not to the number of virtual machines you run on it.
5 host classes · 16 to 384 physical cores per host · up to 1 TB memory
0205Access boundary
Platform operations are included in the host price and are not billed separately: we keep the hypervisor, BIOS and firmware, fabric switches and storage controllers up to date. Above this begins your side — guest system, middleware, application, data. If you want to move the boundary upwards, you book an operations level above it; it does not replace platform operations, but builds on top of them.
Platform operations included in the host price · initial response P1 30 minutes · P1 around the clock, P2 and P3 Mon–Fri 08:00–18:00
0305Oversubscription
How many virtual cores can be placed on a physical core is fixed per host class. Placement does not exceed this value, not even briefly and not even if a neighbouring host is currently quiet. Memory is not oversubscribed at all in the production classes: ballooning — the subsequent withdrawal of memory from running guests — costs more latency under load than the gained memory is worth.
PC-M64: 6 vCPU per core, 0% memory overcommit · PC-X256: 4 vCPU per core
0405Separation
Datastores are attached to a dedicated storage fabric, guests to the vRack. Separate paths are the reason why a nightly replication does not shift the response times of a database — and at the same time the reason why a class without a separate fabric must not run shared datastores.
PC-S32: 3 Gbit/s uplink without separate storage fabric — shared datastores are not permitted on this class, and without a shared datastore there is no automatic restart after a host failure.
0505Maintenance
Firmware and hypervisor maintenance requires a host to be emptied beforehand. In a cluster, the guests migrate away and no one notices the window. On a single host, the same window means downtime — this is the second reason for the minimum cluster size, alongside automatic restart.
ENTRONYX Managed vSphere: Wed 02:00–06:00 CET, rolling host by host, announcement 14 days in advance · lead time for changes 48 hours, in the “Platform and guest operations” level 24 hours
We keep up to date
You keep up to date
Firmware maintenance requires a host evacuation. Without a second host in the cluster, this means downtime of up to 45 minutes per maintenance window.
Escalation: NOC › Platform engineer on duty › Infrastructure on-call management
Management, hosts, storage, vRack — this is the order in which a Private Cloud is built, and the order in which it falls apart in the event of a fault. The circuit diagram shows case ENTRONYX Managed vSphere; the other 3 stacks differ in the minimum number of hosts the second tier carries.
Topology
Schematic, not to scale · 2–32 hosts
Architecture of a cluster under ENTRONYX Managed vSphere
Storage and guest traffic deliberately run on separate paths: datastores are attached to the fabric, guests to the vRack. Classes without a separate fabric — such as PC-S32 with a 2 Gbit/s uplink — are therefore not approved for shared datastores.
Cluster layers under ENTRONYX Managed vSphere with the limits guaranteed in the catalogue. All amounts net.
| Layer | Component | Guarantee and limit |
|---|---|---|
| 01 · Management | ENTRONYX Managed vSphere | Dedicated vCenter per customer, API access, SSO against your directory |
| Maintenance windows | Wed 02:00–06:00 CET, rolling host by host, announcement 14 days in advance | |
| Licensing | €9.90$11.48 per core and month, minimum purchase 16 cores per socket | |
| 02 · hosts | Cluster size | minimum 2, maximum 32 hosts |
| Permitted classes | PC-M64 · PC-M64i · PC-L128 · PC-X256 | |
| vCPU commitment | Upper limit of the respective host class, never more generous | |
| 03 · Storage | Datastore NVMe Performance | 120,000 IOPS · 0.4 ms p99 · €0.1090$0.1264 per GB and month · from 2 TB · Triply replicated across two fire zones of the location |
| Datastore NVMe Standard | 45,000 IOPS · 0.9 ms p99 · €0.0720$0.0835 per GB and month · from 1 TB · Triply replicated across two fire zones of the location | |
| Datastore Capacity | 9,000 IOPS · 4.0 ms p99 · €0.0290$0.0336 per GB and month · from 8 TB · Erasure coding 8+2 in the location, cache triply mirrored | |
| Datastore Archive | 900 IOPS · 25.0 ms p99 · €0.0068$0.0079 per GB and month · from 20 TB · Erasure coding 8+2, second copy in another EU region | |
| 04 · vRack | vRack 1 Gbit/s | included in the host price — Isolated Layer 2 network across all locations. |
| vRack 10 Gbit/s | €39.00$45.24 per month — For storage replication and database clusters. | |
| vRack 25 Gbit/s | €119.00$138.04 per month — Dedicated fabric for HPC and hypervisor clusters. | |
| vRack 50 Gbit/s | €219.83$255.00 per month — For distributed storage pools whose write path runs over the private network. | |
| vRack 100 Gbit/s | €421.49$488.93 per month — Highest expansion stage for clusters with synchronous replication between fire compartments. |
Limitations
In the Public Cloud, the boundary between two customers runs in the hypervisor of a host that both occupy. This is a robust boundary — it is certified, tested and supports regular operations. However, it is a boundary in software.
In the Private Cloud, it lies one tier lower: a host does not carry any third-party guests because none are assigned to it. How you isolate virtual machines from each other on it is up to you — via separate network segments, roles and, for stacks with their own management tier, via a management instance that belongs exclusively to you.
As far as network separation is concerned, responsibility remains shared: we provide the segmentation and rule framework, you set the specific rules. This row is therefore further up in the matrix under “shared” and not under our name.
This overview answers whether the Private Cloud is the right fit. What needs to be decided next — stack, size class, storage, recovery tier — is detailed on its own page with the complete limits. The metric on each card tells you whether the journey there is worth it.
What HA (automatic restart), DRS (load balancing) and vMotion (moving running machines) actually do in operations — and what licensing per core costs compared to licensing per socket.
How much raw capacity remains after mirroring or erasure coding (distribution with parity blocks instead of a full copy) and from which cluster size the protection mode changes.
Which HANA size class fits which dataset, how the volume layout is created and what an acceptance report objects to.
Who patches which layer, when escalation happens and what lead time a change needs — level by level.
Which IOPS and latency commitment applies per class, under which condition it applies and from what quantity it is billed.
Which recovery point (RPO) and recovery time (RTO) each level guarantees, how far away the target region must be and what the recovery test includes.
ENTRONYX HCI and ENTRONYX Managed vSphere share the same host classes and the same network connection; they differ in where the storage is located. If you haven't decided yet, you will find the calculation for this on both pages — they list the same classes with a different licence rate.
The price of a managed host is made up of the bare metal, platform operations and the licence for the chosen stack. How many virtual machines run on it does not change this — the limit is set by the host's cores and memory, not by a count of the guests.
The range spans from €510.59$592.28 to €11,130.78$12,911.70 per host and month: at the bottom PC-S32 under ENTRONYX KVM Platform, at the top PC-X256 under ENTRONYX HCI. Between both ends lie 16 versus 384 physical cores and two licence models that handle cores differently.
Both amounts apply per host. ENTRONYX Managed vSphere requires 2 hosts, ENTRONYX HCI 3 — not as a sales rule, but because an automatic restart needs a second host and the distributed storage pool needs a quorum, i.e. a majority of reachable hosts that decides on the valid data state. The table therefore lists both for each stack.
Not included in the host price are datastores (from €0.0068$0.0079 per GB and month, up to €0.1090$0.1264), disaster recovery and the operations levels above platform operations. Platform operations itself costs nothing extra — it is already included in the host price.
All amounts net. The entry level corresponds to €607.60$704.81 incl. 19% VAT.
The complete calculation including datastore, recovery level and term is in the configurator. It calculates with the same rates as this table.
| Stack | Licence | Hosts | Smallest class | Per host from | Minimum cluster from | Configurator or request |
|---|---|---|---|---|---|---|
| ENTRONYX KVM PlatformOpen hypervisor on libvirt, licensed per socket instead of per core. | €49.00$56.84per socket and month | 1 – 16 | PC-S3216 cores · 5 classes approved | €510.59$592.28 | €510.59$592.281 × PC-S32 | Open |
| ENTRONYX Managed vSpherevCenter-managed cluster with HA, DRS and vMotion — licence per core. | €9.90$11.48per core and month | 2 – 32 | PC-M6448 cores · 4 classes approved | €1,498.20$1,737.91 | €2,996.40$3,475.822 × PC-M64 | Open |
| ENTRONYX HCIHyperconverged: computing power and storage in the same host, licence per core. | €13.90$16.12per core and month | 3 – 24 | PC-M6448 cores · 4 classes approved | €1,690.20$1,960.63 | €5,070.60$5,881.903 × PC-M64 | Open |
| ENTRONYX SAP CertifiedCertified HANA nodes, licensed per host — without overcommitment. | €1,290.00$1,496.40per host and month | 1 – 8 | PC-M6448 cores · 4 classes approved | €2,313.00$2,683.08 | €2,313.00$2,683.081 × PC-M64 | Request |
The licence depends on the stack, not the host: with licensing by cores, a consolidation host pays the same as two smaller ones with the same number of cores; with licensing by sockets, it halves the licence costs. Which model is cheaper for your setup therefore depends on the class — the calculation for this is on the two stack pages.
A managed host costs more than a server you operate yourself. The difference is not a markup on the hardware: it consists of platform operations and licence. The box breaks down PC-M64 under ENTRONYX Managed vSphere into these items — using the amounts from the same catalogue that the configurator uses.
In the host price of PC-M64, the machine itself is listed at €844.00$979.04 net per month — Basic configuration, without hypervisor, licence and updates. The surcharge for the managed host is not a markup on the hardware, but storage expansion, platform operations and licence.
Proof: Catalogue: hardware item PC-M64, €844.00$979.04 net
The host is on the balance sheet at €1,023.00$1,186.68, the licence next to it at €9.90$11.48 per core and month — you see what the Operations cost, and you see what the licence costs. What is included in the host price is broken down by the box “What our amount consists of” into bare metal, storage expansion and platform operations.
Proof: Licence per core and month openly disclosed
Neither a fee per cluster nor a one-off setup fee. The host price is the price; the initial response for P1 in 30 minutes is included.
Proof: Catalogue: setupNet 0, platform operations 0 €$ additional
We only bill managed hosts per month. If you need computing power for a few days, the Public Cloud is the right choice: an hourly rate applies there.
Proof: Catalogue: host classes with monthly amount
A trial here is not a separate offer but the smallest contract: one host, a one-month term, the same rates as in production. What applies is set out in four lines — with the section cited wherever it comes from the Terms and Conditions.
One host with shared datastore
1 × PC-M64 · Managed dedicated hosts
48 cores · 232 GB allocatable · Frankfurt am Main
Datastore NVMe Standard, 1 TB
Monthly net
€1,145.73$1,329.05
An automatic restart after a host failure only takes effect from the second host. The shared datastore, which it also requires, is already included in this entry-level option.
The smallest class with local datastores
1 × PC-S32 · Managed dedicated hosts
16 cores · 140 GB allocatable · Frankfurt am Main
Local datastores in the host, included in the host price
Monthly net
€510.59$592.28
PC-S32 has no separate storage fabric and therefore does not support a shared datastore — an automatic restart after a host failure is not available in this class, even with a second host.
How a contract ends and how long a price holds are governed by the Terms and Conditions in the same way for all product lines — including the Private Cloud. Here are four sentences on each, with the section where the binding wording can be found.
Exact wording: § 11 Term and termination · § 12 Backup and data return
Exact wording: § 8 Prices and price changes
Summarised. The wording of the Terms and Conditions is authoritative.
What distinguishes the Private Cloud from the Public Cloud and bare metal — each with the value by which it can be verified, and the condition under which it applies.
A managed host carries exclusively your guests. The upper limit for vCPU per physical core is an upper limit and not a target value — nothing is placed above it, not even briefly.
Ballooning is excluded in the production classes: what is allocated to a guest is physically available. The gained memory would not be worth the latency it costs under load.
Firmware, BIOS, hypervisor, fabric switches and storage controllers are part of platform operations, and this is included in every managed host — not bookable as an additional tier.
Each datastore class carries an upper limit for the p99 latency — the response time that 99 out of 100 accesses fall below — and the condition under which it applies. Where the condition does not apply, the worse value is shown next to it.
Synchronous mirroring acknowledges every write operation at both locations. It is limited to a location pair, mirrors logical errors and therefore does not replace a backup.
For certified SAP nodes, an acceptance protocol is available per node. Limit value and measured value stand side by side; any change to RAM or datastore suspends the commitment until the new protocol.
No names, no quotes, no logos. Each case states the initial situation, the chosen setup — including the question of why it did not belong in the Public Cloud — and a verifiable metric.
Manufacturing
Initial situation
A HANA production system must run against an acceptance protocol. In the Public Cloud, this commitment cannot be made: the certification depends on a named configuration, and shared capacity changes it with every placement.
Decision
ENTRONYX SAP Certified on a dedicated node, vCPU one-to-one with the physical core, separate volumes for data, log and shared. The price of this specification is included in the contract: any change to memory or datastore suspends the commitment until a new acceptance protocol is issued.
Result
310 µs
Log write latency, 4 KB in the acceptance run, limit ≤ 1000 µs
Public administration
Initial situation
A procedure must not lose an acknowledged write operation if a location fails. The Public Cloud was ruled out here because the storage of individual blocks cannot be committed to a specific location there.
Decision
ENTRONYX Managed vSphere across two locations with level “Synchronous metro mirroring”. Every write operation waits for acknowledgement from the second location, and logical errors are mirrored as well — the level complements the daily backup, it does not replace it.
Result
0 minutes
Guaranteed recovery point, Synchronous metro mirroring
Software and services
Initial situation
An application platform runs around the clock at approximately the same load and grows in steps. In the Public Cloud, this base load is paid per allocated vCPU (€16.69$19.36 per vCPU and month), in the Private Cloud per physical core (€31.21$36.20) — and the cluster grows in hosts, not in instance sizes.
Decision
ENTRONYX HCI from 6 nodes: computing and storage are in the same device, the cluster grows by individual nodes. Below this size, the pool runs mirrored instead of encoded — this costs usable capacity.
Result
62%
Usable share of raw capacity from 6 nodes, erasure coding 4+2
Questions that are actually asked in sales and operations. Where a commitment has a condition, the condition is in the same answer.
The stack decides this, not the size of your environment. ENTRONYX KVM Platform runs from 1 host, ENTRONYX Managed vSphere requires 2, ENTRONYX HCI 3.
The reason is the same in all three cases: an automatic restart after a host failure requires a second host to take it over, and a datastore that both can see. Below this limit, there is no high availability — not even on a small scale.
No. A managed host is billed monthly because it is assigned as a device and does not come from a pool. Switching it off does not reduce the amount.
If you size for the daily peak, you therefore pay for it at night and at the weekend too. For daily fluctuating loads, the Public Cloud is the right fit; the “Cross-check” section on this page explains the case in plain text.
Varies per stack. ENTRONYX KVM Platform bills per socket and month, ENTRONYX Managed vSphere per core and month, ENTRONYX HCI also per core.
When licensing by cores, a minimum purchase per socket applies: below 16 cores, this number is still charged. Small sockets are therefore not worthwhile under such a model.
How many virtual machines you run on a host does not change the licence. The limit is set by the cores and memory of the host.
Yes. ENTRONYX Managed vSphere provisions a vCenter instance in your tenant, not a shared one. Access includes the API and authentication against your directory.
What is not possible with this: a custom-built hypervisor state, your own kernel below the guest, or a firmware of your choice. These layers are our responsibility — that is the core of the offering and at the same time its limit.
In platform operations, which is included in every managed host, we patch Hypervisor, BIOS and firmware, Fabric switches, Storage controllers. Guest system, middleware and application remain with you.
If you want to push the boundary further up, book a higher operations level; there are 3. Guest patches are only applied to the images we deliver there — custom-built images remain your responsibility.
Not in the production classes. For PC-M64 and all larger classes, the permitted memory overbooking is 0%: what is allocated is physically available.
An upper limit per physical core applies to vCPUs — for PC-M64 it is 6. It is an upper limit and not a target value: nothing is placed above it, not even briefly.
The entry-level class is exempt from this; it is intended for development and staging environments and explicitly allows overbooking.
Announced and rolling. For ENTRONYX Managed vSphere: Wed 02:00–06:00 CET, rolling host by host, announcement 14 days in advance. The hosts are evacuated one after the other, not together.
Firmware maintenance requires that a host can be emptied. Without a second host in the cluster, this means downtime per maintenance window — this is the second reason for the minimum cluster, alongside the automatic restart.
That depends on the booked level. The highest, Synchronous metro mirroring, acknowledges every write operation at both locations and thus guarantees a recovery point of 0 minutes and a recovery time of 20 minutes.
The level has two conditions that are part of the commitment: it is limited to a location pair at metro distance, and it mirrors logical errors as well. It complements the backup, it does not replace it.
At least one host class is available at 4 locations in Germany. Not every class is available everywhere: the most strictly managed class is available at 3 locations.
Datastore classes and recovery levels have their own location lists. Synchronous mirroring requires two locations at metro distance and is therefore not available everywhere.
The configurator guides you from the platform via the host class to the datastore, the network and the operations level including recovery. It calculates with the same rates that are on this page, and shows the monthly invoice before ordering — including the minimum cluster size that the chosen stack enforces. Each initial situation above opens it already filled out.