parallelvirtualcluster
/
pvc
1
0
Fork 0
Code Issues 11 Pull Requests Projects Releases 111 Wiki Activity

Move information about memory utilization

This commit is contained in:
Joshua Boniface 2020-12-21 00:16:49 -05:00
parent 78c017d51d
commit d22a5aa7f2
1 changed files with 12 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
docs/cluster-architecture.md
View File

@ -46,8 +46,6 @@ The following table provides bare-minimum, recommended, and optimal specificatio
| Total RAM (n-1) | 32GB | 96GB | 128GB | | Total RAM (n-1) | 32GB | 96GB | 128GB |
| Total disk space | 200GB | 400GB | 800GB | | Total disk space | 200GB | 400GB | 800GB |
Of these totals, some amount of CPU and RAM will be used by the storage subsystem and the PVC daemons themselves, meaning that the total available for virtual machines is slightly less. Generally, each OSD data disk will consume 1 vCPU at load and 1-2GB RAM, so nodes should be sized not only according to the VM workload, but the number of storage disks per node. Additionally the coordinator databases will use additional RAM and CPU resources of up to 1-4GB per node, though there is generally little need to spec coordinators any larger than non-coordinator nodes and the VM automatic node selection process will take used RAM into account by default.
### System Disks ### System Disks
The system disk(s) chosen are important to consider, especially for coordinators. Ideally, an SSD, or two SSDs in RAID-1/mirroring are recommended for system disks. This helps ensure optimal performance for the system (e.g. swap space) and PVC components such as databases as well as the Ceph caches. The system disk(s) chosen are important to consider, especially for coordinators. Ideally, an SSD, or two SSDs in RAID-1/mirroring are recommended for system disks. This helps ensure optimal performance for the system (e.g. swap space) and PVC components such as databases as well as the Ceph caches.
@ -62,6 +60,18 @@ The general rule for available resource capacity planning can be though of as "1
For memory provisioning of VMs, PVC will warn the administrator, via a Degraded cluster state, if the "n-1" RAM quantity is exceeded by the total maximum allocation of all running VMs. This situation can be worked around with sufficient swap space on nodes to ensure there is overflow, however the warning cannot be overridden. If nodes are of mismatched sizes, the "n-1" RAM quantity is calculated by removing (one of) the largest node in the cluster and adding the remaining nodes' RAM counts together. For memory provisioning of VMs, PVC will warn the administrator, via a Degraded cluster state, if the "n-1" RAM quantity is exceeded by the total maximum allocation of all running VMs. This situation can be worked around with sufficient swap space on nodes to ensure there is overflow, however the warning cannot be overridden. If nodes are of mismatched sizes, the "n-1" RAM quantity is calculated by removing (one of) the largest node in the cluster and adding the remaining nodes' RAM counts together.
### System Memory Utilization
By default, several components of PVC outside of VMs will have large memory allocations, most notably Ceph OSD processes and Zookeeper database processes. These processes should be considered when selecting the RAM allocation of nodes, and adjusted in the Ansible `group_vars` if lower defaults are required.
#### Ceph OSD processes
By default, PVC will allow several GB (up to 4-6GB) of RAM allocation per OSD to maximize the available cache space and hence disk performance. This can be lowered as far as 939MB should the administrator require due to a low RAM configuration, but no further due to Ceph limitations; therefore at least 1GB of memory per storage OSD is required even in the most limited case.
#### Zookeeper processes
By default, the Java heap and stack sizes are set to 256MB and 512MB respectively, yieliding a memory usage of 500+MB after serveral days or weeks of uptime. This can be lowered to 32M or less for lightly-used clusters should the administrator require due to a low RAM configuration.
### Operating System and Architecture ### Operating System and Architecture
As an underlying OS, only Debian GNU/Linux 10.x "Buster" is supported by PVC. This is the operating system installed by the PVC [node installer](https://github.com/parallelvirtualcluster/pvc-installer) and expected by the PVC [Ansible configuration system](https://github.com/parallelvirtualcluster/pvc-ansible). Ubuntu or other Debian-derived distributions may work, but are not officially supported. PVC also makes use of a custom repository to provide the PVC software and an updated version of Ceph beyond what is available in the base operating system, and this is only compatible officially with Debian 10 "Buster". PVC will, in the future, upgrade to future versions of Debian based on their release schedule and testing; releases may be skipped for official support if required. As a general rule, using the current versions of the official node installer and Ansible repository is the preferred and only supported method for deploying PVC. As an underlying OS, only Debian GNU/Linux 10.x "Buster" is supported by PVC. This is the operating system installed by the PVC [node installer](https://github.com/parallelvirtualcluster/pvc-installer) and expected by the PVC [Ansible configuration system](https://github.com/parallelvirtualcluster/pvc-ansible). Ubuntu or other Debian-derived distributions may work, but are not officially supported. PVC also makes use of a custom repository to provide the PVC software and an updated version of Ceph beyond what is available in the base operating system, and this is only compatible officially with Debian 10 "Buster". PVC will, in the future, upgrade to future versions of Debian based on their release schedule and testing; releases may be skipped for official support if required. As a general rule, using the current versions of the official node installer and Ansible repository is the preferred and only supported method for deploying PVC.