Pick the right pool for the job

[depasseg]

b. more vdevs=more iops and more bandwidth

 

[Black Ninja]

b. more vdevs=more iops and more bandwidth

I haven't test it , but I believe you are correct.:)

Here is much harder one for you: I have currently 2 pools in my signature freenas. Can you tell , which one is faster* ?


*faster - consider one larger file sequential read

 

[depasseg]

Same logic applies. more vdevs=faster=RAID 10 (even though that's not the real name, but I don't know what it's really called).

And since you have 2 pools, why not do a test to see?

 

[Bidule0hm]

I'd say b. too ;)

And for the challenge: I'd say the RAID-Z2 (if we are talking about a local read (no network) and that the ARC doesn't impact the result of course) :)

 

[Black Ninja]

Same logic applies. more vdevs=faster=RAID 10 (even though that's not the real name, but I don't know what it's really called).

And since you have 2 pools, why not do a test to see?

I already test it. This was more like a brain teaser challenge, and I am afraid you are wrong. It was my supprice for me too to find out that raidz2 was faster.

 

[gpsguy]

I would have guessed b too.

It sounds like Black Ninja has already done the test.

For me it's good to have 2 pools with different raid levels for different purposes. Example: one pool for storage one for speed and you put you data accordingly

 

[Black Ninja]

I'd say b. too ;)

And for the challenge: I'd say the RAID-Z2 (if we are talking about a local read (no network) and that the ARC doesn't impact the result of course) :)

You are correct for both. But I wonder if you just guess the challenge question or you know why.

By the way disregard ARC , and I mean network, not local. What good is local (synthetic) speed results for NAS device ?

 

[Black Ninja]

I would have guessed b too.

It sounds like Black Ninja has already done the test.

I did with the hdd I have as of now, But I havent tried other combination(1 pool with 16 identical drives)

 

[Bidule0hm]

It was my supprice for me too to find out that raidz2 was faster.

But I wonder if you just guess the challenge question or you know why.

It's not a big surprise (and I just thinked logically) because it's a big sequential read, not a lot of random small reads. With the striped mirrors you'll have more IO/s because you have more vdevs, but you'll have less throughput because you've less drives with about the same unitary throughput (the 5.9 k drives aren't a lot slower than the 7.2 k ones because what matters most is the seeks delays).

In short:

• vdevs gives IO/s
• drives gives MB/s

What good is local (synthetic) speed results for NAS device ?

About half a GB/s start to be good. Some configs can tickle the GB/s (based on the famous dd tests with bs=2048k count=50k).

 

[depasseg]

I've got to throw a coach's challenge. Take a look at these test results. I think something might have skewed your results.

http://zfsguru.com/forum/buildingyourownzfsserver/570#5242 (the x-axis in the graphs below is # of drives).

 

[depasseg]

It's not a big surprise (and I just thinked logically) because it's a big sequential read, not a lot of random small reads. With the striped mirrors you'll have more IO/s because you have more vdevs, but you'll have less throughput because you've less drives with about the same unitary throughput (the 5.9 k drives aren't a lot slower than the 7.2 k ones because what matters most is the seeks delays).

In short:

• vdevs gives IO/s
• drives gives MB/s

About half a GB/s start to be good. Some configs can tickle the GB/s.

Except in a RAID Z2 - the checksum data has to be read from all the drives greatly reducing the sequential read aspect.

 

[joeschmuck]

What good is local (synthetic) speed results for NAS device ?

That is an easy one, I've used this before and it's pretty slick. Intel NAS Performance Toolkit (NASPT).

 

[Black Ninja]

It's not a big surprise (and I just thinked logically) because it's a big sequential read, not a lot of random small reads. With the striped mirrors you'll have more IO/s because you have more vdevs, but you'll have less throughput because you've less drives with about the same unitary throughput (the 5.9 k drives aren't a lot slower than the 7.2 k ones because what matters most is the seeks delays).

In short:

• vdevs gives IO/s
• drives gives MB/s

It's you are right on all counts. As a matter of fact if make my second pool raid0 with all 6 drives striped for speed it's still slower than my first pool of 10xRaidz2 because of the drives.
Those 2TB 7200rpm drives as individual drives(made year 2010) can read about 110MB /s compared to 4TB 5900Rpm(made year 2014) who can read as single drive about 140MB/s. I assume seek times are better like you said , and older ones are 512 bytes sector vs 4k sectors.

 

[Black Ninja]

That is an easy one, I've used this before and it's pretty slick. Intel NAS Performance Toolkit (NASPT).

I will try it

 

[Black Ninja]

I've got to throw a coach's challenge. :) Take a look at these test results. I think something might have skewed your results.

http://zfsguru.com/forum/buildingyourownzfsserver/570#5242 (the x-axis in the graphs below is # of drives).


View attachment 7873

View attachment 7874

View attachment 7875

View attachment 7876

Very good info. I love these. There was this web site https://calomel.org had very good test info , but the web is gone for some reason.

 

[Black Ninja]

Except in a RAID Z2 - the checksum data has to be read from all the drives greatly reducing the sequential read aspect.

It's still faster. The difference comes because of the difference in the drives , even in raid0 they are slower than Raidz2 in first pool.

 

[jgreco]

Probably the two vdev setup.

 

[depasseg]

It's you are right on all counts. As a matter of fact if make my second pool raid0 with all 6 drives striped for speed it's still slower than my first pool of 10xRaidz2 because of the drives.
Those 2TB 7200rpm drives as individual drives(made year 2010) can read about 110MB /s compared to 4TB 5900Rpm(made year 2014) who can read as single drive about 140MB/s. I assume seek times are better like you said , and older ones are 512 bytes sector vs 4k sectors.

There is something with your drives that is causing a major slowdown. There is no way that a 6-disk 7200rpm stripe should be slower than a 10-disk 5400rpm RAID Z2.

How are you testing? do you have results for: 10 disk stripe, 10 disk RAID-Z2, 6 disk stripe, 6 disk RAID-Z2.

 

[Black Ninja]

There is something with your drives that is causing a major slowdown. There is no way that a 6-disk 7200rpm stripe should be slower than a 10-disk 5400rpm RAID Z2.

How are you testing? do you have results for: 10 disk stripe, 10 disk RAID-Z2, 6 disk stripe, 6 disk RAID-Z2.

I don't remember the results , but remember that on 7200 rpms from raid10 to raid0 it wasn't much faster. Maybe the seek times play major role, or since that are older , what is some of them is slowing the other by reading trouble , even they are 100% smart helaty.

 

[depasseg]

RAID 10 vs RAID 0 with 6 disks should be pretty similar. RAID Z2 OTOH should be noticeably different.