Low end system: ISCSI/Windows read speed slower than write speed.

[n8lbv]

Hi,
I'm still brand new to Freenas and this is NOT a production system.
It's just for testing and learning on.
Actually quite pleased with the performance speed for just testing.

1GB isolated test LAN, write speed is on par with the actual write speed of the magnetic drive.
Read is slower and I am curious why.

During read/write test,
RAM and CPU usage on the dashboard look relatively low. nothing is being maxed out.

Anyhow trying to get a better understanding of the system and what would typically cause read speed to be down compared to write speed
When I am quite used to the opposite being the case.


Dell Optiplex 390 *Latest Motherboard Firmware installed.
Service TAG: 8TX38V1
CPU Gen2 i5
RAM 8GB DDR3
test pool: Two WD Black WD1003FZEX in a stripe.
Using built on motherboard Intel SATA ports
FreeNAS-11.2-U5

 

[n8lbv]

I'm getting same/similar speed when testing an iscsi share on a twin SSD stripe.

 

[n8lbv]

Any ideas or pointers why the read speed seems capped around 80?
And SSDs X2 drive themselves are a lot faster but not seeing any of the speed on ISCSI.

 

[jgreco]

You have 8GB. There is no space for read caching.

Writes are going to be faster.

You need about 64GB as an entry point for iSCSI. Even then, reads tend to be slower because reads are responsive (i.e. filer has to get a request, go out to SSD, retrieve it, then send it back over the wire) whereas writes can be cached in RAM if you're not running async.

https://www.ixsystems.com/community...res-more-resources-for-the-same-result.28178/

https://www.ixsystems.com/community...xi-nfs-so-slow-and-why-is-iscsi-faster.12506/

etc

 

[n8lbv]

Thanks!!
Can't wait to try this on real hardware.

 

[n8lbv]

I'll bug you with one more question and then leave you alone.
Given the correct hardware, correct amount of RAM and all.
Is Freenas a good/great enterprise ISCSI high performance solution compared to others?
I will give it a shot on 64GB+ of ram just for fun to see how it plays.

 

[jgreco]

The main problem with ZFS is that is a resource hog. ZFS gains performance from computer science trickery that exploits resource richness to cover for the mechanical limitations of HDD.

As a copy-on-write filesystem, ZFS will be REALLY fast writing data as long as there are huge amounts of contiguous free space available on your pool. In practice, if you keep your pool utilization somewhere less than 40%, you can mistake the write performance for that of an SSD based system. (This is a complex topic.)

Copy-on-write also means that fragmentation is a potentially severe issue. ZFS leverages the ARC and L2ARC mechanisms to address that. In most cases, the ways people use iSCSI are such that there's a working set ("hot data" or whatever) that is frequently accessed and cold data that is rarely or never accessed. If you can size the system ARC and L2ARC to cover the working set, it's possible to get to a point where you are almost never reading anything off the pool disks and are instead fulfilling 99%+ read operations out of ARC or L2ARC.

This has been a major selling point for ZFS for many years, because it means that you can get SSD-like performance out of a ZFS filer.

However, flash prices have been falling for some years now, and we're near a tipping point where the math on this may change in the next 5 years or so.