Skip to content


You launching a live server, which is most probably connected to a network, so a little planning is definitely useful. If you just need a machine to test your web-based app or a place to run your command line tools, the defaults should be 'OK'. If you need multiple instances or unusual ports, contact before beginning, unless you really know what you are doing.

Logging into the cloud

The cloud URL is . To get access, start a new project or gain access to an existing one please contact UTHPC support by emailing .

Log in with your UT credentials, the domain name is

In case you would prefer to use the self-service , please ask your Project Lead to invite you to the project or request access here .

  1. If using University's central username and password, use as the domain.
  2. University central user name if domain is ’’.
  3. University central password if domain is ’’.

Starting a new instance

Choose ’Instances’ from the left-hand menu and click the ’Launch Instance’ button.

On every page there is a small help icon with ?-sign that provides you with further information about the available options.


  1. Instance name without dots or non-ASCII chars. is automatically added.
  2. Availability zone – only one zone available anyway.
  3. How many instances to create.

From now on, you don't have to fill all values, but a few are critical. The default values are usually 'OK'.


Choose the boot source. If you select ’Image’ then you can choose several Linuxes, BSD, and Windows Images. Otherwise – it depends.


Flavors are features for virtual machine (RAM,CPU/GPU storage etc.) Getting GPU's requires access to GPU flavors. Same applies to bare metal nodes.

Flavors define underlying hardware size and type. Flavors are mostly formatted as follows type<generation>.label

m and c – general purpose memory or compute optimized flavors.

  • m1 and c1 are first generation hardware from 2013.
  • m2 and c2 are second generation hardware from 2018 and are approx. 35% faster than the first generation.
  • m3 and c3 are third generation hardware from 2020 and are approx. 25% faster than the second generation.
  • m4 and c4 are fourth generation hardware from 2023 and are approx. 37% faster than the third generation.

GPU – flavor with appropriate GPU accelerator.

bare metal – entire server with everything that's inside.

Resizing ’normal’ virtual machines without PCI passthrough (GPU) works fine on Linux VM's for example moving back-and-forth between older m1/c1, m2/c2 flavors and newer m3/c3.


Networks – for network access

  1. University internal network provider_64 – is accessible only from internal University networks or VPN, no firewall in-house, except for the security rules you define in the UI and inside the machine.
  2. Network inside of cloud. You can add floating IP from provider network to make it available from outside.
  3. Network accessible from the outside world publicly - provider_11, provider_33, and provider_154. TCP ports 22, 80, 443, and ICMP are public by default. You must enable those also in your security group if needed. If there is a need for other ports, you must ask the UTHPC team to open them. Also, please think whether you really need your machine to be accessible for the whole world before using a public address.
  4. public_ipv6_net – this one is public /64 ipv6 network range only for you project. You must use dualstack IPv4 and IPV6 networks because metadata (your ssh keys , hostname etc) is IPv4 only.


If you need *.[hpc.] DNS then launch your instance into private network and allocate ’Floating IP’ from appropriate provider network. For that please see ’Network Topology’ tab. Then send allocated IP together with required name to . By default you get automatically <instance_name> if the name is already not taken.

There are daily hacking tries against machines in public networks, sometimes with success. Make sure you have:

  • turned off password authentication,
  • applied tight firewall policy,
  • installed software from trusted sources.


Compromised machines are instantly shut off.

Network ports

No need to add any. Please just move on by clicking ’Next’.

Security groups

Here you can configure firewall rules.

  1. Predefined firewall rules – you can choose zero or many. Make tight rules if you are using public network segment. Also, you can not mix instance wide rules and port specific rules. It's needed to choose one approach or another. If needed create a new group, for example HTTP and allow ingress TCP 80 and 443.

Key pair

You definitely need a key pair to access your instance – in fact it's the only way to get in.

  1. Hit ’Create Key Pair’ button and follow instructions.

You need RSA public/private key pair. Please save your private key and keep it safe. Your private key is only loaded on first boot. You can add extra keys manually by modifying ~/.ssh/authorized_keys file.

Now you can click ’Launch Instance’.

Access your server

Sooner or later you need web console access because the machine won't boot or networking is down etc. Local user password is only for web console and not for SSH. SSH with password is by default turned off. Adding password for local user still means that SSH password authentication is still turned off. Web console is black area below:

Should you need help with the basics, for example how to set up SSH connection, please refer to the SSH manual .

Please keep in mind

  • Ubuntu-based image default user is ubuntu,
  • Debian-based image default user is debian,
  • CentOS-based image default user is centos,
  • Flatcar-based image default user is core,
  • Rocky-based image default user is rocky,
  • FreeBSD-based image default user is freebsd,
  • Windows-based image default user is Admin.

To get password execute nova get-password and access it like this: rdesktop -u Admin -p supersecret

Another option is to be first one who logs in.

You can always define your private network and associate floating IP from provider network. This is a recommended way to use persistent IP's. There must be a virtual router between provider and private network. You can define it. If you want some resolvable host name, then you need persistent IP.

Private overlay network expected speed inside cloud is at least 1 Gb/s. Provider network expected speed is 3-20 Gb/s, but it depends of flavor generation and network setup.

OpenStack firewall concept name is ’security group’. You must review/create rules if you need some extra port/protocol enabled. You can apply many security group rules to VM.

Available images

  • centos7 - Red Hat community supported long-term support.
  • centos9 - CentOS Stream with watchdog device.
  • rocky8 - Rocky Linux 8.
  • rocky9 - Rocky Linux 9.
  • flatcar - Flatcar Container Linux.
  • ubuntu22.04 – Ubuntu 22.04 LTS (Jammy Jellyfish) with watchdog device.
  • ubuntu20.04 – Ubuntu Server 20.04 LTS (Focal Fossa).
  • debian11 – Debian 11 (Bullseye).
  • debian12 – Debian 12 (Bookworm) with watchdog device.
  • freebsd12.0 – FreeBSD 12.
  • freebsd13.0 – FreeBSD 13.
  • WS2019 – Windows Server 2019 with GUI, only for UT staff and students. Request access by e-mail and include project name or UUID.

Creating your own image based on previous images are also possible and probably the easiest way to bake your passwords or special "whatever" in. Those images are by default private and only visible within your project. Ubuntu 22.04, Debian 12 and CentOS Stream 9 also has a watchdog device '/dev/watchdog'. You can add your own custom monitoring scripts to '/etc/watchdog.d/'

Docker installation inside VM

Please avoid docker default bridge docker0 network This network is already taken and you lose your connection as soon as bridge is coming up.

You can run command ip route get your_laptop_ip to predict if you are loosing connectivity when docker is coming up. If Your IP is within docker0 range you loose connectivity. Please make sure that you have local sudo user with password. This makes web console usable.

To understand how to setup Docker so that your network is fully functional, please see the setting up docker in ETAIS page


  • Launch instance from image named openstackclient-major-minor.
  • Via web interface push following button. V3 is a good option.
  • Compute -> Access and Security -> API access -> Download OpenStack RC.
  • And source downloaded file inside launched instance.


Available models are: ’A100’, ’P100’, ’Tesla K40c’, and ’Tesla K40m’.

VM's with GPU's are hard wired with hypervisor where they're initially booted. If hypervisor dies then your VM with GPU dies.

Bare metal nodes

If needed then it's also possible provision bare metal, as opposed to virtual, machines. Bare metal flavor represents physical server in server room. Everything that's within metal enclosure. In order to get access to bare metal flavors and networks please send a support request to .

Flavor with name baremetal.dl165 gives root access to hardware with following spec:

  • ProLiant DL165 G7
  • 2x AMD Opteron(TM) Processor 6276
  • 128 GB of RAM
  • 1x MB0500EBNCR 500 GB HDD, 1xCrucial_CT2050MX300SSD1 SSD
  • 2x Quad Data Rate (QDR) internal InfiniBand (10G Ethernet overlay) and 1x 1 GB external Ethernet

Flavor with name baremetal.vedur gives root access to hardware with following spec:

  • ProLiant DL165 G7
  • 2x AMD Opteron(TM) Processor 6276
  • 192 GB of RAM
  • 1x MB0500EBNCR 500 GB HDD
  • 2x Quad Data Rate (QDR) internal InfiniBand (10G Ethernet overlay) and 1x 1 GB external Ethernet

Notes for booting bare metal instances ( ):

  • Source: baremetal
  • Flavor: baremetal
  • Networks: baremetal
  • Security Groups: ‘remove all allocated SG’

Notes for network:

  • By default InfiniBand is in datagram mode. It's possible to change it to ’connected’ mode to get better Ethernet bandwidth over InfiniBand by running command: root@baremetal-node:~# echo “connected” > /sys/class/net/<intf name>/mode.

Object storage

S3 and Swift API on top of 'prod3' storage backend. The cloud URL is

  • get openrc (OpenStack RC) and source your environment
  • run command openstack --os-interface public ec2 credentials create
  • install boto3
  • use API endpoint

Temporary Linux containers

Temporary Linux containers are available at .

  • Boot time around 10 seconds.
  • Container lifetime 24 h, automatic cleanup.
  • SSH authentication – public key only, web console – password only.
  • Root access inside container.
  • Container is centos7 based.
  • Firewall block from container to UT internal non-routable networks.
  • Shared 256G RAM, 32 cores, 1T storage.
  • No backup.
  • No abuse limits.
  • No SLA – if it dies then it dies.


  • prod storage type – weekly tape backup, snapshots and volumes. Running VM backup may be inconsistent.
  • prod2 storage type – weekly tape backup, snapshots and volumes. Running VM backup may be inconsistent.
  • prod3 storage type – nightly tape backup, snapshots only.

Storage performance

Performance depends of disk capacity. Bigger disks are faster. Roughly 1 TiB disk characteristics are below:

  • prod type storage (shared production HDD, performance depends of global load):

    • 1 GB/s, after 10 seconds limited to 600 MB/s
    • 600 IOPS, after 10 seconds limited 300 IOPS
  • prod2 type storage (dedicated production HDD):

    • 1.5 GB/s, after 10 seconds limited to 600 MB/s
    • 3000 IOPS, after 10 seconds limited 2000 IOPS
  • prod3 type storage (IOPS intensive SSD):

    • 1.5 GB/s, after 30 seconds limited to 600 MB/s
    • 20000 IOPS, after 30 seconds limited 10000 IOPS

E-mail,, relayhost is Message transfer agent (MTA), is rate limited.