Quick hardware question

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Guentha

New Member
Sep 21, 2023
6
0
1
I am building my first truenas. I have always used S2D Hyperconverged setup with my Hyper-v cluster but performance is low and S2D has been unreliable and there are not a lot of great tools or knowledge when things go wrong.

I don't want to add a lot of extra hardware to this build but I am open if it makes sense. most of my hardware is free from my local college.

here is what I have:
dell r820, 4x E5-4600 v2 , currently 512GB ram( I have way more 16GB sticks but anything over 512 drops ram speed from 1600 to 1333, I assume I will be better off loading it with 768GB?)

4x 56GB Infiniband connections to redundant switches. all Hyper-V hosts have 4x connections as well.

HBA 9500-16e connected to Hitachi 4u60 DAS

I have 20 1.8TB enterprise SSD I am thinking about L2Arc
I have hundreds of 12TB xenos Sata drives but will only run 14-20 as I don't need the space or the power usage.
20 or so 200gb high endurance Sata SSD for Slog, thinking of using 4 in a dual mirror setup if this is doable. can upgrade to NVME if this will help.

This will be used to serve 4 IBM x3850 X6 hyper-v hosts. each host will have 10-20 virtual servers. most the server don't read or write much but there are a few windows virtual servers that seem to always be chewing through iops.

I am mainly worries about IOPS. I have also seen offloading the Metadata....should I do that? does that need NVME?
 

cheezehead

Active Member
Sep 23, 2012
730
176
43
Midwest, US
All depends on the workload. For TrueNAS the hardware is overkill on the cpu side and possibly the ram depending on how much space or active space you need.

Ram will get used as ARC space (read cache) before L2Arc comes into play. If you have lots of ram, adding L2Arc may not be of any help. SLOG will take the write hit, but again it depends on the actual read/write workload.

Sizing for overall space is pretty simple. Getting numbers for how the read/write io blender is for your workload is going to be the trick. LiveOptics or another tool may give you an idea. As you only have a single box for the Truenas deployment build it with what ram you have laying around then start moving workloads on it to get more familiar with how the caching works for you.