ZFS SLOG not speeding up writes, and recommended size for 10 GBit LAN

flashdrive

Patron
Joined
Apr 2, 2021
Messages
264
Having added a 128 GByte SLOG device (Samsung 830 SSD) in TrueNAS CORE and SCALE I do not see any write speed improvements during 10 GB LAN write operations.

  • How can I make sure the the SSD is being utilised? The IO stat for sure.
  • Is there any tuning to be done?
In my understanding the SLOG device needs to be faster / more responsive then the HDD pool to see a benefit.


SLOG Devices
Disk latency is the primary concern for SLOG devices. The SLOG device only needs to store as much data as the system can throughput over the approximately 5 second “flush” period. With a 1GB connection, this is about 0.625 GiB. A 10GB connection requires about 6.25 GiB and 4x10 GiB requires 25 GiB.

 

c77dk

Patron
Joined
Nov 27, 2019
Messages
468
gstat, zilstat and zpool iostat are what I would look at
 

Ericloewe

Server Wrangler
Moderator
Joined
Feb 15, 2014
Messages
20,194
Do you have sync writes? If you don't, an SLOG is going to gather dust and do little more.

128 GByte SLOG device (Samsung 830 SSD)
Not at all a good choice of SLOG. Since it does not have power loss protection, it is no safer than eschewing sync writes entirely. You'd get the worst of both worlds: no ZIL and worse performance.
Its performance characteristics are also going to be unimpressive for SLOG applications, but that's the least of your concerns.
 

flashdrive

Patron
Joined
Apr 2, 2021
Messages
264
Hello,
thank you for your feedback.

I have found this thread:

My goal for my homelab host / server is the same: SMB writes / ingest mostly.

So what would be the way to go to have a good write performance?

For now I am around 400 Mbyte / s with 5 HDDs and 10 GB LAN.

I know that by writing to a RAM disk I can get near to the theoratical max of 10 GB LAN copper based.
 

rvassar

Guru
Joined
May 2, 2018
Messages
972
Hello,
thank you for your feedback.

I have found this thread:

My goal for my homelab host / server is the same: SMB writes / ingest mostly.

So what would be the way to go to have a good write performance?

For now I am around 400 Mbyte / s with 5 HDDs and 10 GB LAN.

I know that by writing to a RAM disk I can get near to the theoratical max of 10 GB LAN copper based.

SMB isn't going to make use of a SLOG much, if at all.

To maximize performance here, you're probably going to want some kind of multi-vdev pool configuration. A single vdev RAIDz2 pool will have limited write performance. Consider mirrors, or a multi-vdev RAID config where the kernel can round-robin to the other group of disks and perform writes in parallel. But even then, you have to consider the spindle performance of each drive. I have a 4 device mirror pool, and I still can't hit full 10GbE speeds. Since each drive tops out at 150 mb/sec or so, I'm guessing I need 6 or 8 drives in 3 or 4 mirror vdevs.

This is why all those little sub-2Tb 2.5" 10k & 15k rpm drives lingered in the Enterprise for so long. They didn't hold much, but prior to reasonable flash/NVMe, they put up serious IOPS numbers in large groups.
 

Ericloewe

Server Wrangler
Moderator
Joined
Feb 15, 2014
Messages
20,194
Windows does not request many sync writes over SMB. Mac OS seems to do a fair bit more sync writes, though.
 

flashdrive

Patron
Joined
Apr 2, 2021
Messages
264
Windows does not request many sync writes over SMB. Mac OS seems to do a fair bit more sync writes, though.

Ok, however I want Windows OS to send the data to the NAS.
 

flashdrive

Patron
Joined
Apr 2, 2021
Messages
264
Top