# Libvirt Realms

The Libvirt realm uses KVM/QEMU to control the VMs. For this to work, we need:

1. ticrypt-vm service must have root access to all the VM Host servers. This is accomplished by injecting the RSA-2048 public key of ticrypt-vm into the /root/.ssh/authorized_keys file for each of the VM Hosts.

2. A distributed file-system shared by all the VM Hosts must be set up. The storage must be large enough to accommodate the VM Images, VM Drives, and Libvirt temporary files.

Two independent tasks need to be accomplished for the Libvirt realms to function correctly:

• Realm setup

• Hardware description

The realm setup is specified in the file included from the realms section of ticrypt-vm. The hardware description goes into ticrypt-vm config file itself.

# Configuring the Libvirt Realm

note

All the parameters go in the included file specified in the realms section.

caution

These parameters should not be part of sections inside the included file.

ParameterTypeRequiredDescription
disabledBooleanIs the realm disabled?
driverLivbirtThis is a Libvirt realm
nameStringDisplayable name of realm
volumes-poolStringLibvirt name for volumes pool
drives-poolStringLibvirt name for drives pool
bricks-poolStringLibvirt name for VM Images
log-schedulingBoolDebug VM scheduler?
registration-timeoutDurationVM startup timeout
poll-frequencyDurationHow often a VM is checked
network-filterStringDefault network filter
drives.lazy-allocationon, offDelay allocation?
drives.cacheStringCache strategy
drives.min-sizeSizeMinimum drive size
drives.max-sizeSizeMaximum drive size
uploadsSectionReserved for future use
scalingSectionReserved for future use

Some notes on the above are in the following order.

info

volumes-pool, drives-pool and bricks-pool have default values ticrypt-vm, ticrypt-vm-drives and ticrypt-bricks. Unless you are hosting multiple tiCrypt instances, there is no reason to change the defaults

danger

registration-timeout cannot be too short since some VMs might not have time to boot and register within the specified time and would fail to work.

note

network-filter is an obscure feature that needs to be specified only if problems arise. See link: Libvirt Format and Filer for a more complex explanation.

tip

You almost always want drives.lazy-allocation=on. This way, you "pay" for the storage only when it is needed.

• drives.cache is best left at "default". In general, it is tricky to select. Possible values are:
• "default": System default caching strategy, may be either "writethrough" or "writeback" depending on the version of qemu-kvm
• "writethrough": Host page cache is enabled, but disk write cache is disabled, meaning all writes to the drive only complete when the data has been committed to the underlying storage. Similar to adding O_DSYNC flag to writes.
• "writeback": Both the host page cache and disk write cache are enabled. Disk behaves similar to a raid controller with RAM cache.
• "none": Bypasses host page cache entirely. Similar to adding O_DIRECT flag to reads and writes.

# Hardware specification preliminaries

There are four main components in VM host hardware configuration:

Curves: Built-in parameterized functions taking in the value of a single attribute (e.g. number of VMs, amount of memory used) and returning a cost based on that attribute. Used in the construction of Cost Functions.

Cost Functions: Configurable rules that determine how much it costs to schedule resources on a specific node. Used to determine if VMs can be scheduled to specific notes, and which node is most optimal.

Hardware Profiles: Descriptions of a hardware configuration. Link together a cost function along with resource information. Assigned to hosts to allow for scheduling and device allocation.

Hosts: Individual machines used to run client VMs. Hosts are assigned a hardware profile to use for scheduling decisions.

note

Cost Functions and Hardware Profiles are managed in the ticrypt-vm configuration file, while Hosts are dynamically managed using the tiCrypt user interface.

The types of values used in the description of parameters are:

TypeExampleDescription
String"an example"String value
Int42Whole number value
Bytes32, 64 MiB, 18 GiBA number of types with an optional unit
Real1.0, 1A real number with an optional decimal place
RealOrBytes1.0, 32 MiB, 2 GiBEither a real number or several bytes with unit
Curve{curve: invalid}A curve definition
CurveMap{id1: {...}, id2: {...}}A mapping of IDs to curve definitions
Device{type: "gpu", addr: "02:03.4"}Object containing a device’s type and PCI address
DeviceList[{...}, {...}]A List of Device entries

# Curves

Curves are configuration objects with a single required string parameter, curve. This refers to one of the following built-in curves which take a single attribute and compute a cost from it. This may be the number of VMs running, the amount of memory being used in bytes, or the number of specific devices in use. Additional parameters may be specified depending on the curve type.

If a curve has multiple values listed for its types, any of those can be used as the value of the curve attribute.

The following parameters are common to all curve types:

• Common Curve Parameters
ParameterTypeRequiredDefaultDescription
scaleReal1.0The scale factor to apply to the curve result.
maxRealNoneIf specified, any input past this value will return Infinity.

## Constant curve​

Types: constant

Always returns a fixed value.

• Constant Curve Parameters
ParameterTypeRequiredDefaultDescription
valueRealicon:check[role="green"]The constant value to return.
• Constant Curve Examples

{ curve: constant, value: 1.0 }

## Invalid Curve​

Types: invalid

Always returns positive infinity, preventing the VM from being scheduled. Mostly useful in conjunction with the <<Piecewise Curve>>.

• Invalid Curve Examples

{ curve: invalid }

## Linear Curve​

Types: linear

Computes a value from the input based on a line with a given slope and base.

• Linear Curve Parameters
ParameterTypeRequiredDefaultDescription
slopeRealOrBytesicon:check[role="green"]The slope of the line.
baseRealOrBytesicon:check[role="green"]The vaule to return when the input is 0.
• Linear Curve Examples

{ curve: linear, slope: 1.0, base: 0.0 } { curve: linear, slope: 10.0, base: 5.0 }

## Hard-Soft Curve​

Types: hard-soft, soft-hard

Creates a mostly continuous piecewise linear curve with a soft and hard cap. The cost ramps up slowly until the input reaches soft, at which the cost ramps up much more quickly until the input reaches hard. After that, the cost is infinite.

• Hard-Soft Curve Parameters
ParameterTypeRequiredDefaultDescription
softRealOrBytesicon:check[role="green"]The input value at which softAmt should be returned.
hardRealOrBytesicon:check[role="green"]The input value at which hardAmt should be returned.
softAmtReal1.0The cost returned when the input value reaches soft.
hardAmtReal10.0The cost returned when the input value reaches hard.
• Hard-Soft Curve Examples

Returns 0 when input is 0, 1 when input is 1, 10 when input is 2, infinity when input > 2

{ curve: "hard-soft", soft: 1.0, hard: 2.0 }

Both of the following are equivalent:

{ curve: "hard-soft", soft: 32 GiB, hard: 48 GiB, softAmt: 10, hardAmt: 100

{ curve: "hard-soft", soft: 32 GiB, hard: 48 GiB, scale: 10.0 }

## Unavailable Curve​

Types: unavailable

Creates a piecewise curve that returns either 0 if the input is less than or equal to zero, or infinity if the input is greater than zero. This is useful for denoting unavailable resources.

note

This curve is the default for any devices that do not have curves associated with them.

• Unavailable Curve Examples

{ curve: unavailable }

## Piecewise Curve​

Types: piecewise, low-high, high-low

Returns the result of one of two curves depending on the input value. Allows for more complex piecewise linear curves.

• Linear Curve Parameters
ParameterTypeRequiredDefaultDescription
cutoffRealOrBytesicon:check[role="green"]The cutoff point for switching from low to high
lowCurveicon:check[role="green"]The curve to use when the input is less than or equal to cutoff
highCurveicon:check[role="green"]The curve to use when the input is greater than cutoff
• Piecewise Curve Examples

Constant cost of 5 up to a hard cutoff when input is 6

{ curve: piecewise, cutoff: 6.0, low: { curve: constant, value: 5}, high: { curve: invalid } }

Same as a hard-soft with soft = 1, hard = 2, softAmt = 1, hardAmt = 10

{ curve: piecewise cutoff: 1 low: { curve: linear slope: 1 base: 0 } high: { curve: piecewise cutoff: 2 low: { curve: linear slope: 9 base: -8 }, high: { curve: invalid } } }

Same as unavailable:

{ curve: piecewise, cutoff: 0, low: { curve: constant, value: 0 }, high: { curve: invalid } }

# Cost Functions

Types: [[cost-fct]]

Cost Functions are managed in the ticrypt.vm.cost-functions section of ticrypt-vm.conf, which is mapping of cost function IDs to their definition. Cost function definitions have the following structure. See the section <<Curves>> for more information about parameters with the type Curve.

• Cost Function Parameters
ParameterTypeDefaultDescription
nameStringSame as IDA human-readable name for the cost function.
descriptionStringEmptyA description of the cost function.
vmsCurveNoneCurve used to assign cost based on number of VMs running.
vcpusCurveNoneCurve used to assing cost based on number of virtual CPU cores used.
memoryCurveNoneCurve used to assign cost based on memory usage in bytes.
devicesCurveMapNoneMapping of device ID to curve for cost of using devices of that type.
scaleReal1.0Scaling factor to multiply all costs generated by this function.
offsetReal0.0Static offset to add to all costs generated by this function. Not affected by the scale.
note

Any device types not listed in the devices section will act as if they were defined with the unavailable curve.

danger

A curve should be set for at least one of vms, vcpus, or memory; otherwise, there will be no restriction on VMs using no devices being scheduled to hosts using this cost function. This will cause rapid resource exhaustion.

## Cost Function Examples​

Basic Cost Function

ticrypt.vm.cost-functions {    simple {        name: "Simple"        description: "A basic cost function"        # Add 1 cost per VM scheduled to the VM        vms: { curve: "linear", slope: 1, base: 0 }        # VCPUs cost 0.5 each until 8, then 3 each until 16        vcpus: { curve: "hard-soft", soft: 8, hard: 16, softAmt: 4, hardAmt: 32 }        # Add 10 cost at 16 GiB of memory, 100 at 32 GiB        memory: { curve: "hard-soft", soft: 16 GiB, hard: 32 GiB, scale: 10}        # alternative way to specify the same thing:        #memory: { curve: "hard-soft", soft: 16 GiB, hard: 32 Gib, softAmt: 10, hardAmt: 100 } }}

Cost Function with GPUs

ticrypt.vm.cost-functions {    nvidia-gpu {        name: "GPU Node"        description: "Used for GPU nodes with the standard configuration"        # Same as simple        vms: { curve: "linear", slope: 1, base: 0 }        vcpus: { curve: "hard-soft", soft: 8, hard: 16, softAmt: 4, hardAmt: 32 }        memory: { curve: "hard-soft", soft: 16 GiB, hard: 32 GiB, scale: 10}        # NVidia GPUs cost 1 each, up to a max of 2 used        devices.nvidia-gpu: { curve: linear, slope: 1, base: 0, max: 2}        # Extremely high offset so that VMs do not get scheduled here unless there is no        # other choice        offset: 100000    }}

Cost Functions Using Pools and Includes

.ticrypt-vm.conf

ticrypt.vm.cost-functions {    pool1 {        include "cost-functions/pool-common.conf"        name: "Pool 1"    }    pool2 {        include "cost-functions/pool-common.conf"        name: "Pool 2"        # Add a flat 1000 so that hosts in pool 2 will not be used unless the VM could not be        # scheduled to hosts in pool 1        offset: 1000    }}

cost-functions/pool-common.conf

• vms: { curve: "linear", slope: 1, base: 0 }

• vcpus: { curve: "hard-soft", soft: 8, hard: 16, softAmt: 4, hardAmt: 32 }

• memory: { curve: "hard-soft", soft: 16 GiB, hard: 32 GiB, scale: 10}

# Hardware Profiles

Types: [[hardware-profile]]

Hardware Profiles are defined in the ticrypt.vm.hardware-profiles section in ticrypt-vm.conf. Each Hardware profile should be defined as an ID mapped to a configuration block with the following parameters:

• Hardware Profile Parameters
ParameterTypeRequiredDefaultDescription
nameStringSame as IDA human-readable name for the hardware profile.
descriptionStringEmptyA description of the hardware profile.
coresInticon:check[role="green"]The number of CPU cores available on the host.
memoryBytesicon:check[role="green"]The amount of memory in bytes available on the host.
devicesDeviceList[]The PCI devices available for VM use.
cost-functionStringicon:check[role="green"]The ID of the cost function to use for this profile.

## Hardware Profile Examples​

ticrypt.vm.hardware-profiles pool1: {        name: "Pool 1"        description: "Standard host in pool 1"        cores: 8        memory: 32 GiB        cost-function: "pool1"    }    pool2: {        name: "Pool 2"        description: "Standard host in pool 2"        cores: 8        memory: 32 GiB        cost-function: "pool2"    }        nvidia-gpu-node: {        name: "NVidia GPU Node"        description: "Special nodes with NVidia GPUs attached"        cores: 8        memory: 32 GiB        cost-function: nvidia-gpu        devices = [            {type: "nvidia-gpu", addr: "0000:01:02.0"}            # PCI addresses can omit the domain if it is all 0            {type: "nvidia-gpu", addr: "02:03.0"},        ]    }

# Hosts

Hosts are managed from the Management tab in the tiCrypt user interface. Administrators are able to add, enable, disable, and remove hosts from this interface.

Each host's configuration consists of the following pieces of information:

Name
A descriptive name of the specific host.

URI
The Libvirt connection URI used to connect to the host. Full documentation on the various URI formats are available at the Libvirt documentation.

Some examples include:

• Local QEMU Connection: qemu:///system

• Remote QEMU Connection: qemu+ssh://[user@]host[:port]/system

Hardware Profile
One of the defined hardware profiles to use for resource and scheduling information.

State
The operational state of the host. Can be one of the following:

• Enabled: A connection will be established to the host, and VMs will be scheduled to it.
• No Scheduling: A connection will be established to the host, but no new VMs will be scheduled to it. This allows for the management of existing VMs without allowing new VMs. Useful for nodes that need to be brought down gracefully.
• Disabled: No connection to the host will be made, and no VMs will be scheduled to it.

If you used the recommended default values for volumes-pool, drives-pool and bricks-pool, there are four pools that need to be set up in libvirt: ticrypt-vm, ticrypt-vm-drives, ticrypt-bricks and ticrypt-vm-snapshot.
In order to define pools, e.g. ticrypt-vm-snapshots, you can:
virsh pool-define-as ticrypt-vm-snapshots dir - - - - "<path/to/pool>"virsh pool-build ticrypt-vm-snapshotsvirsh pool-start ticrypt-vm-snapshotsvirsh pool-autostart ticrypt-vm-snapshots