Hi, I have been asked to prepare a Truenas Scale server with unusual characteristics.
It will be used by a small group of software developers to host throw-away windows/linux virtual machines for testing purposes (and possibly docker apps, same purpose).
What makes it unusual is that data integrity is not of paramount interest (its just for testing, no real data would be lost), but disk speed of the VMs is.
Here is the hardware I have (I can change it if deemed necessary):
cpu: Ryzen 7 5700g (8 cores, 16 threads)
mb: MSI MEG B550 UNIFY
ram: 128GB (no ecc)
nvme: 1x Samsung 980 Pro 500GB
ssd: 2x Kingston A400 240GB
hdd: 1x Seagate Barracuda 2TB (7200rpm)
software is Truenas Scale 22.02.1
The best I could obtain (single datapool on the fastest nvme disk) is this (tested from inside a win10 virtual machine):
For reference, same vm with aggressive PrimoCache settings (2GB ram cache, deferred intelligent writes 300s max latency) gives me this:
Which is of course far better.
Is there any way to configure zfs/truenas to use a similarly aggressive caching scheme (I mean large read cache and deferred writes)?
As I already said, we are aiming for maximum io speed here, data safety is not a priority.
Thanks,
Carlo
It will be used by a small group of software developers to host throw-away windows/linux virtual machines for testing purposes (and possibly docker apps, same purpose).
What makes it unusual is that data integrity is not of paramount interest (its just for testing, no real data would be lost), but disk speed of the VMs is.
Here is the hardware I have (I can change it if deemed necessary):
cpu: Ryzen 7 5700g (8 cores, 16 threads)
mb: MSI MEG B550 UNIFY
ram: 128GB (no ecc)
nvme: 1x Samsung 980 Pro 500GB
ssd: 2x Kingston A400 240GB
hdd: 1x Seagate Barracuda 2TB (7200rpm)
software is Truenas Scale 22.02.1
The best I could obtain (single datapool on the fastest nvme disk) is this (tested from inside a win10 virtual machine):
For reference, same vm with aggressive PrimoCache settings (2GB ram cache, deferred intelligent writes 300s max latency) gives me this:
Which is of course far better.
Is there any way to configure zfs/truenas to use a similarly aggressive caching scheme (I mean large read cache and deferred writes)?
As I already said, we are aiming for maximum io speed here, data safety is not a priority.
Thanks,
Carlo