BUILD Supermicro 5018A-MHN4 build

[Martin Renters]

If anyone is interested in a 1U Intel Atom C2758 build with four 4TB WD Red drives, here's a write-up of what I did and the performance of the system.

http://www.teckelworks.com/2014/10/building-a-freenas-server-with-a-supermicro-5018a-mhn4/

Martin

 

[Ericloewe]

It's an interesting choice. Don't you find the 1U format somewhat limited?

 

[Martin Renters]

Limited in what way? This is for a home server for image storage and some backup. It has close to 8TB of usable space with relatively low power consumption. I can always upgrade the drives to 6TB versions for 12TB of usable space. I plan on getting a second unit to put in a data center for off-site replication and some VM storage.

Martin

 

[areis]

I have a similar setup used for photo storage (Aperture) and Time Machine backup. However, I went with the Supermicro X10SLM-F and a E3 1240 Xeon because I also run Plex. My FreeNAS server works awesome!

This community has been a great help.

• Read the guides and howtos before you post.
• Accept constructive criticism.
• Have thick skin.

Good luck.

 

[mjws00]

Looks like a mini in a 1U form factor to me. With Supermicro vs AsRock and 2578. Nice.

The benchmarks and conclusions based on them are off. Anything with compression is pretty much measuring cpu and bus speeds. You need very large datasets to ensure you've blown out the ARC as well. The ~388MBps numbers (bonnie++, read, compression off) is pretty reflective of your pool and matches common knowledge. I'm a little surprised to see such disparity between your read vs. write and suspect it is testing methodology vs actual performance. If you hunt you'll see the numbers often land at 10-20% or less reads vs writes but much of that is workload dependent. Doesn't matter much as even dual 1GBe is easily saturated.

Nice pictures and site. Welcome.

 

[cyberjock]

The benchmarks and conclusions based on them are off. Anything with compression is pretty much measuring cpu and bus speeds. You need very large datasets to ensure you've blown out the ARC as well. The ~388MBps numbers (bonnie++, read, compression off) is pretty reflective of your pool and matches common knowledge. I'm a little surprised to see such disparity between your read vs. write and suspect it is testing methodology vs actual performance. If you hunt you'll see the numbers often land at 10-20% or less reads vs writes but much of that is workload dependent. Doesn't matter much as even dual 1GBe is easily saturated.

You are correct. The results are obviously useless. Benchmarking ZFS is different than benchmarking other file systems and the same methodologies just don't work. I'm sure if someone is interested there's plenty of other people that have tried to use bonnie++ and made some mistakes that make the numbers meaningless and some of it was explained to them. Benchmarking ZFS isn't for amateurs. If you don't know what you are doing you can get numbers that seem somewhat realistic that you can put into charts and graphs and think it means something, but really doesn't mean much more than what you could get from /dev/random.

 

[Martin Renters]

I'm a little surprised to see such disparity between your read vs. write and suspect it is testing methodology vs actual performance. If you hunt you'll see the numbers often land at 10-20% or less reads vs writes but much of that is workload dependent.

I'm not sure I understand your comment. What is it you are surprised about?

Martin

 

[mjws00]

I'm not sure I understand your comment. What is it you are surprised about?
Martin

I was referring to your writes being so slow vs. your reads. However, while I was reading I was thinking of a z1 or z2 array and not your striped mirrors. Got a little thrown by the odd numbers, sorry. The ratio is not interesting.