The Compute Cloud is the Infrastructure-as-a-Service (IaaS) offering from LRZ. It can provide individual virtual machines (VMs) to its customers to run HPC workloads or to provide machines needed for testing purposes or to cover short-running computational tasks.
The service will be provided following the best effort principle. It is not suited to host services that need to be highly available. It is highly recommended to use the "Managed Server" offering for these kind of services.
In this context, "best effort" also means that users cannot be given a guarantee to access a particular type of hardware (like GPU nodes) at a specific time. There is no reservation mechanism available. Furthermore, we cannot offer guarantees to create new VMs at any time, e.g. if the cloud resources are fully occupied by other users.
Compute resources will be made available to all users in the same fashion without prioritising particular users.
The LRZ Attended Cloud Housing offering provides a possibility to invest into own hardware, that can be bought through LRZ; if users want to have exclusive access to computational resources they can buy a server fitting their needs that will be integrated into the Compute Cloud and operated by LRZ. Access to these resources will be limited exclusively to their owners. For more information please refer to chapter "Attended Compute Cloud Housing".
Access to the cloud is offered via a web portal and APIs. Using these interfaces users can create and manage their VMs and infrastructure like networks and storage following a self-service approach. The users have privileged (root) access to their machines and can configure them to fit their needs. That being said, the users themselves are responsible to keep their VMs up to date and to make sure that they own the appropriate software licenses (if needed).
Flavors
Flavors are used to assign phyiscal ressources to a particular VM. Each flavor defines on which hardware a VM will run and how many CPU cores, the amount of main memory, ... are used (a complete pyhsical host or a fraction). It is not possible to provide individually configured flavors.
By default users to not have access to all kinds of flavors. Access is limited to the basic flavors providing access to up to 10 CPU cores. If users need access to more resources or GPUs they need to contact the cloud support team that can increase the particular user's quota and provide access to the requested flavors.
Quotas
Quotas are used to limit the number of resources that can be used concurrently by a user.
The use of quoata prevents that the cloud resources can be completely occupied by a single user maliciously or by accident preventing VMs of other users to get started. It also creates awareness that the physical resources are limited.
That being said, quotas can be adjusted by the cloud support team, if there is a valid reason for that.
The default quota should match the needs for most users:
This means that a user can start up to 4 VMs running simultaneously that can together use up to 10 virtual CPU cores. For example, there might be some smaller VMs (e.g. 2x lrz.large + 1x lrz.medium) or one large VM with 10 cores (lrz.xlarge).
If a user needs more resources these quotas can be raised upon request by the project's master user, i.e. if a GPU should be used in on of the VMs.
The quota cannot be exceeded by the user; new machines will not be started or volumes will not be created if this would exceed the current user quota.
Deletion of VMs and data
The LRZ reserves the right to delete VMs and data from projects and users if they were deactivated or deleted more than three months ago or if their budget has been used up for three months.
Users are responsible for backing up their data.
Scope of Services and Service Specifics
Management
LRZ offers self-service access to management functionality of the Compute Cloud via a web interface or the API for
- virtual machines (VMs)
- operating system images
- data volumes including snapshots
- networks
Virtual Servers
As already mentioned the compute cloud offers different flavors to assign certain amounts of ressources to a VM. It either provides access to a full physical nodes or a fraction of it.
We use CPU overprovisioning in the LRZ Compute Cloud to satisfy the needs of as many customers as possible. Main memory will not be overprovisioned.
Operating systems and software
LRZ offers operating system images for multiple Linux distributions as well as FreeBSD that are compatible with the Compute Cloud infrastructure. These images are provided directly by the distributors and are not altered by LRZ. For more information please refer to docs.openstack.org/image-guide/obtain-images.html.
Users can also create own operating system images that must to be compatible to the x64_64 architecture of the hardware.
The user is responsible for having the correct licenses needed by the installed operating system or any other software used by the VMs.
The LRZ Compute Cloud is an Infrastructure-as-a-Service (IaaS) offering. We cannot support users with regard to their individually needed configuration or software within the VMs.
Storage
LRZ offers block-based storage used for operating system images and volumes for the VMs.
The storage offers redundancy to keep data available in case of a hardware failure. However, LRZ does not create backups.
The user can create snapshots and clones but this does not replace backups each user should create individually.
If the user needs more than 1 TiB storage capacity the usage of different cloud storage offerings is highly recommended, e.g. Cloud Storage or Data Science Storage (DSS) for capacity exceeding 20 TiB.
Network
Cloud users can use the virtual networks provided by LRZ or create own virtual networks.
Security Groups isolate the own virtual machines from any other VMs running in the cloud. Security Groups can be seen as virtual firewalls built in the cloud infrastructure. They can be fully configured by the user.
Additional firewalls are not put in place by LRZ: Access from and to the virtual infrastructure is not restricted by LRZ.
To get access to the VMs the LRZ Compute Cloud uses the concept of floating IPs. These can either belong to MWN-available IPv4 address space or to a worldwide accessible IPv4 address space.
Capacity planning
Access to resources is limited per user. This limit can be adjusted on request via the service desk by a master user.
LRZ cannot and does not guarantee access to a certain amount or type of resources at a given point in time. The cloud is operated on a best-effort basis.
Fees
The use of the LRZ Compute Cloud is subject to a charge, depending on the resources used.
When creating a VM, the user specifies the flavour with which it is to be created. Among other things, the flavour regulates the allocation of resources to the VM, i.e. the number of CPU cores, the size of the main memory and whether a GPU is provided.
The LRZ provides the following flavours for the use of customer VMs, which have different prices depending on the resources provided. A complete list of all available flavours can be found on the documentation platform.
| Name | vCPUs | RAM | GPUs | (S) per Day | (V) per Day | (S) per Year | (V) per Year |
|---|---|---|---|---|---|---|---|
| lrz.tiny | 1 | 1,12 GiB | 0 | on Demand | on Demand | on Demand | on Demand |
| lrz.small | 1 | 4,5 GiB | 0 | on Demand | on Demand | on Demand | on Demand |
| lrz.medium | 2 | 9 GiB | 0 | on Demand | on Demand | on Demand | on Demand |
| lrz.large | 4 | 18 GiB | 0 | on Demand | on Demand | on Demand | on Demand |
| lrz.xlarge | 10 | 45 GiB | 0 | on Demand | on Demand | on Demand | on Demand |
| lrz.2xlarge | 20 | 90 GiB | 0 | on Demand | on Demand | on Demand | on Demand |
| lrz.4xlarge | 40 | 180 GiB | 0 | on Demand | on Demand | on Demand | on Demand |
| nvidia-v100.1 | 20 | 368 GiB | 1 | on Demand | on Demand | on Demand | on Demand |
Costs are only charged for virtual machines to which resources are allocated. If a VM is switched off (‘shelved’), no costs are incurred for it.
Budgets
In order to limit the maximum costs that can be incurred, there are two types of budgets that are managed by the master user:
1. project budget: the project budget sets the upper limit for the total costs that can be generated by the project. The total costs are made up of the costs generated by the individual users of the project.
2. user budget: The user budget offers the option of limiting the costs that an individual user can generate.
As soon as a budget is exceeded, the relevant VMs are switched off (shelved). If the project budget is exceeded, all VMs of the project are shelved; if a user budget is exceeded, all VMs of the corresponding user are shelved.
VMs can only be restarted (i.e. their data can only be accessed again) once the budget has been increased.
The LRZ reserves the right to delete VMs and data if budgets have been exceeded for more than three months. Users are responsible for backing up data in good time.