So you want to buy a brand new server to run virtual machines on it. Here are some advices to achieve maximum perfomance and how to save some money.
Initial Hardware Setup¶
RAM is the main issue. Multiply the number of concurrent workstations by the amount of memory each one requires and that is the total RAM the server must have. However, recent virtualization improvements allow you to overcommit the memory.
For remote virtual desktops it is enough to have 1 GB ethernet cards. If you are planning on having many video intensive workstations at the same time it would be good to have 10 GB network cards on the host.
You will need to store 3 different kind of data in the server:
Operative System Disk drives¶
The Operative System partitions like root, /usr and /var are critical so the server keeps running. But perfomance is not the main issue. Buy two or three small hard disk drives. Create RAID1 or RAID5 and define these partitions there. In our experience RAID1 is more than enough, if you can afford it, buy 3 disk drives: 2 in the RAID and the third one as a spare.
There is no much space requirement for the operative system. 50 GB should be than enough. If you buy larger disks you may create a partition to store some virtual machines volumes.
With Ravada virtual machines usually are cloned from a base. This base is prepared in advance with all the sofware the users need. All the machines will read information from the base volumes so it is a good idea to store this data in SSD disk drives.
This kind of disks are expensive, so you likely would want to buy only one or two small disk drives for the base volumes.
RAID5 is usually slow so it is not adviced. If you want to have redundancy configure a RAID1 with 2 SSD disk drives for the base volumes. It would usually be mounted at /var/lib/libvirt/images
If you want to save some money do not use RAID for the base volumes. In our experience, top hardware vendor brand disk drives are reliable. You may get more space if you buy 2 SSD drives and create two different partitions. In this case you will have base volumes stored in /var/lib/libvirt/images.1 and /va/lib/libvirt/images.2 .
Be aware that without RAID there is a downtime risk. If one of the disk drives fail the information it contains may be lost and a backup must be restored in a new replacement drive. But it is uncommon to have both disk drives failed at once, so you can restore the data in the other volume and carry on while the replacement arrives.
Clones volumes are incremental information stored on top of base volumes. Usually this data doesn’t require as much perfomance. So it is not a bad idea to save some money here and store the clones volumes in large mechanical disk drives. Anyway if you can afford it buy fast disks for a better user experience.
Really cheap server
Operative System: 2 x Hard Disk Drives 100 GB in RAID1
Volumes: 1 x Solid State Disk drives
Budget perfomance server
Operative System: 2 x HDD 100 GB in RAID1
Bases Volumes: 1 x SSD 500 GB
Clones Volumes: 1 x HDD 1 TB
From this example you can grow as long as your budget allows it. Having more drives may give you more space. If you need high availability 24x7 you have to duplicate the volumes disk drives and set RAID1.
High Availability and performance server
Operative System: 2 x HDD 100 GB in RAID1
Bases Volumes: 2 x SSD 500 GB in RAID1
Clones Volumes: 2 x HDD 1 TB in RAID1
Growing and Scaling¶
Ravada is easy to grow. Start with a tight budget, but try to buy the faster drives you can afford.
When more users start virtual machines at the same time the server may run out of memory. Adding more RAM will give you more concurrent users.
If you run out of disk space you can buy more disk drives and add more storage pools. You can configure some bases go to one storage, others to another. If you buy more storage it can be defined the new clones will be created in new partitions.
Ravada does scale: A cluster can be created from a main server and nodes can be added to it. You can use older hardware, even PCs. Ravada will automatically balance virtual machines and start in the less used nodes. You don’t need shared storage in the clusters, but if you use it start up and clone times will be much faster.
Borg backup is a good free choice, its main advantage is it has good deduplication features.