Nappit AIO - Typical CPU Usage

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

kyo77

New Member
Jul 26, 2016
21
3
3
47
Fairly new to Nappit and ZFS, just setup a AIO box for testing. Created a SMB share on the VM, and initiated a copy of a large file to the share and ESXi is showing about 50% cpu usage. Seems pretty high no? Or this is typical for Xeon D-1541?

Current Setup
ESXi 6.7 with latest patches
MB: D1541D4U-2T8R
RAM: 32GB
CPU: Xeon D-1541
4x WD 250GB drives (old drives only using for testing) in raid-z2
Slog - vmdisk on optane 900p
Network: 1gbs only
Enabled Jumbo frames on VMswitch 9000 MTU

AIO VM
24GB Ram
8 vCPU
VMXNET 3 network adapter
Installed current VMtools
Tuned network for jumbo frames (napp-it // webbased ZFS NAS/SAN appliance for OmniOS, OpenIndiana, Solaris and Linux : Handbücher -Tuning)

Also created a VM on the NFS datastore and ran passmark diskmark and server is showing 50% cpu usage.
 

gea

Well-Known Member
Dec 31, 2010
3,141
1,182
113
DE
A server has to work as fast as possible. As long as there is data to processing or calculating, the cpu load will increase. Only when the cpu is at 100% it will limit further performance improvements ex when the disk subsystem would be able to deliver more data otherwise.
 

kyo77

New Member
Jul 26, 2016
21
3
3
47
A server has to work as fast as possible. As long as there is data to processing or calculating, the cpu load will increase. Only when the cpu is at 100% it will limit further performance improvements ex when the disk subsystem would be able to deliver more data otherwise.
Yeah I guess I didn't expect saturating 1gbs network would utilize 50% of the cpu. Just using some spare hardware around to test, to size up for a potential homelab server. With 50% utilization I think that the Xeon D-1541 might be under powered a bit.

I tested bare metal a few months earlier and I remember seeing better performance, I'm probably going to test that again to do a comparison. I read some of your post and it seemed like there wasn't a huge difference to baremetal vs AIO so I'm just wondering if there is some configuration issue going on.
 

gea

Well-Known Member
Dec 31, 2010
3,141
1,182
113
DE
Xeon 1541 is best when power saving aspect is important. You can build a quite performant AiO server on it with a storage VM and some other VMs with a low power consumption even with 10G links external and internal over vmxnet3 with maybe 5G achievable.

OmniOS is one of the smallest OS options for a full ZFS featured storage server set so its load for ESXi is quite low. Its approach is a just enogh OS for a full featured but very stable iSCSI, NFS and SMB storage appliance, no extras,

btw
I would reduce number of vCPU. Start with one or two for storage. If you give all CPUs to a single VM you generate extra load for the scheduler (ESXi needs CPU as well)
 
Last edited:

Rand__

Well-Known Member
Mar 6, 2014
6,626
1,767
113
Also it used to be the case that the VM has to wait until the configured number of cores are available before getting a slot.
I.e. if you have 4 vcpus assigned, only when 4 are actually free (getting blocked until they are) the vm gets them; so if you are heavily over-committed this can cause significant delays.

Not sure this still applies though
 

asche

New Member
Oct 6, 2017
18
4
3
46
4x WD 250GB drives (old drives only using for testing) in raid-z2
Parity calculations shouldn't hit the CPU that much, but perhaps try with a RAID10? Deduplication disabled? Also, what gea/Rand_ said - reduce vCPUs to 4.
 

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
I can’t speak for your specific environment but as a general rule with what I use it’s 1-core of CPU per 1G network traffic, so a system using virtualized network at 10G speed could need about 10-core to be able to do that.
 

Rand__

Well-Known Member
Mar 6, 2014
6,626
1,767
113
but that only would apply if you have 10 concurrent users wouldn't it;)
 

kyo77

New Member
Jul 26, 2016
21
3
3
47
Thanks for all the advice guys, I will try to use less vCPU. Initially I struggled a bit to configure this so I was concerned it was some sort of configuration error which it might of been. I reconfigured and tested against bare metal which performed way better so made me think about just making it a pure storage server. Then I tried AIO again this time building the VM from scratch I got better CPU utilization about 35% vs 50% with same configuration.
 

kyo77

New Member
Jul 26, 2016
21
3
3
47
Xeon 1541 is best when power saving aspect is important. You can build a quite performant AiO server on it with a storage VM and some other VMs with a low power consumption even with 10G links external and internal over vmxnet3 with maybe 5G achievable.

OmniOS is one of the smallest OS options for a full ZFS featured storage server set so its load for ESXi is quite low. Its approach is a just enogh OS for a full featured but very stable iSCSI, NFS and SMB storage appliance, no extras,

btw
I would reduce number of vCPU. Start with one or two for storage. If you give all CPUs to a single VM you generate extra load for the scheduler (ESXi needs CPU as well)
Yeah I got this board off a member and it was just sitting around, figured I would use it for testing/sizing. Looking to replace my current setup, so trying to determine if it's better to sell this board and get a e5-26xx v3 or this would be sufficient for my needs.