Session by Frank Denneman and Valetin Hamburger
The room is packed, 5 min to start. If WIFI permits, I’ll try to do some live blogging again.
11.26
And that concludes Franks and Valentines session on DRS.
Thank you for viewing
11.20
DPM is part of DRS and its scheduler, in fact, there is no DPM without DRS
- affinity rules whcih apply to DRS also apply to DPM (“Should” vs “Must”)
- based on resource entitlement DPM will set a host to standby or wake up
DPM looks at the load on 40Min timeframes and it uses a 5min boot up time for a host in its calculations
11.18
Since a “Must ron on” prevents a VM running on other hosts, a HA of the VM could fail because of this
“Must run on” rules will remain even after you disable DRS on cluster level !!!
11.16
DRS Affinity and anti-affinity rules
- can be used to prevent/enforce running a pair of VMs on one single host
- can be used to prevent/enforce running a group of VMs on a group of hosts
“Must run on” setting, can’t be violated, not even when an HA event is in progress
“Should run on” can be violated if a HA event is in progress or DRS/DPM is in danger to lose proper functioning.
11.14
Resource pools limits can also limit the resource usage:
- limits are boundaries for resources
- limit cannot be expanded
- limits will enforce resource pools to use only a certain amount of resources
- limits can/will create contention within a resource pool
- limits should only be used for a certain use case.
USE RESOURCE POOL LIMITS WITH CARE !!! They can be harmful to your environment.
11.12
Debunking myth: Popular belief is that a RP-level reservation is a limit (It’s not)
Reservation sets the minimum available physical memory for the resource pool
11.09
Consider the following fact for resource pool memory reservation:
- VMkernel reserves memory to run any virtual machine (memory overhead per VM)
- virtual machine overhead is a reservation and in non-shareable by design
- include this during consolidation calculations and RP level sizing
In vSphere5 the memory overhead per VM has been reduced considerably
11.06
Resource pool level reservations distributed amongst children:
- dynamic entitlement is key for the actual reservation allocation
- reservation allocation differs due to utilization
11.04
How to estimate the right amount of shares per resource pool in 4 steps:
1- Match defined performance SLA’s to resources pools
2- Try to make up a shares per VM model, which is than applied to a resource pool
3- Based on the number of vCPU’s, set the shares on the resource pools
4- Introduce a scheduled task which sets the shares per resource pool, based on the number of VMs/vCPUs they contain (Powershell script at http://www.Yellowbricks.com
11.00
Another tip from Frank: Use 1 level of resource pools, maybe 2 levels but definitely don’t go any deeper.
10.59
Resource pools by default behave as a 16GB / 4 CPU VM
10.58
Resource pools:
(WARNING) Using resource pools as a folder structure. This has some risks involved. Even if not configured, RPs are influencing resource entitlement (that’s what they build for)
Resource pool shares:
Share values are relative: Compare shares values to siblings
Shares are set at VM-level and RP-level
Never create a VM outside / next to a resource pool on the same level
10.53
DRS is NOT looking for a balance of resources on a host !!! You could have 20 VMs on one host and just 2 VMs on another host and still be “balanced”.
10.52
In an environment with DRS:
- the DRS introduces a global scheduler
- DRS scheduler computes ideal resource entitlement as if a cluster was one giant host
- global scheduler calculates resource entitlement of child objects
- local ESXi scheduler is responsible for allocating resources to it’s child objects
10.50
In an environment without DRS:
Conention:
- Will affect all VMs on a single host
- Can only be solved by manually reducing workloads
- May be influenced by setting shares/reservations or even limits
10.48
Each ESXi host contains a CPU and memory scheduler:
- computes resource entitlement for powered-on virtual machines
- admission control deciedes if VMs can be powered on
- “Noisy neighbor” will influence all other VMs on the host
10.46
Memory limits:
- apply even when there are enough resources
- limits the use of available physical resources
Limits are often more harmful than helpful:
- limits are transparent to Guest-OS
- Guest-OS and applications size is acting/caching according to configured memory
- limits can have tremendous impact on performance
- reducing configured CPU/Memory is a far better approach than applying
- if a memory limit is active, ESXi provides the additional consumed memory from the VMs vSwap file (VM swapping occurs)
- When a CPU limit is active, ESXi de-schedules the VM even if CPU resources are available
10.43
Tip from Frank; Monitor memory usage of your VM before setting reservations and thereby denying memory to other VMs
10.42
Frank explaining how shares change per VM when adding more VMs.
Valentin continues with CPU reservations. Explaining about CPU guarantees and influences of shares on admission control.
Remember that CPU reservation doesn’t claim the CPU when VM is idle (is refundable)
Frank about Memory reservation.
Memory reservation is as bad as often referenced: non-refundable once allocated. Windows is zeroing out every bit of memory during startup.
Memory reservation caveats:
- will drop the consolidation ratio
- may waste resources (idle memory can’t be reclaimed)
- introduces higher complexity (capacity planning)
10.36
Live without DRS
VM-level shares
- there defined shares level exists: Low, normal, high
- share values are initially related to virtual hardware configurations. Related to its siblings, diluted while adding more VMs
10.35
Contention:
Short-term contention: usually caused by load correlation, load synchronicity (for example logon hours), brown outs (virus scheduled at noon), Patch day
Long-term contention: Massive consolidation ratio’s, hardware limits exceeded, massive over commitment
10.33
Frank opening this session and we’re starting with a little intro. This session focuses on DRS scheduling and all of its components. Start with vSphere resource basics.
Some basics on resources of a VM. CPU allocation and RAM usage. A virtual Machines resource entitlement is a dynamic task within an ESXi host. It’s influenced by particular parameters.
- Dynamic entitlement: Active use of CPU and memory, level of contention
- Static entitlement: Shares, reservations and limits.
