Anybody knows how to explain bug 5289?

[solarisguy]

https://bugs.freenas.org/issues/5289 In an LZ4 world width of RAID groups is irrelevant​

Description:
There's no need to warn any more about suboptimal RAIDZ configurations in the volume manager.​

Anybody who could explain?

 

[enemy85]

https://bugs.freenas.org/issues/5289 In an LZ4 world width of RAID groups is irrelevant​

Description:
There's no need to warn any more about suboptimal RAIDZ configurations in the volume manager.​

Anybody who could explain?

I think could be related to this
http://blog.delphix.com/matt/2014/06/06/zfs-stripe-width/
But don't know how sure it is

 

[dlavigne]

It is.

 

[c32767a]

Nice article..
Good to have some technical analysis to confirm what I thought I was seeing in my testing.

I will note that in testing large vdevs, CPU and RAM make a huge difference. Eventually I/O becomes CPU limited because of the parity calculations required.. And we already know ZFS loves RAM. :)

 

[cyberjock]

I'm not buying the thought process for this at all. From what I've read there's a very small subset of users that this applies to. But for the vast majority of us (>99%) this means nothing. I've yet to see anything to dispute these claims and there's very little detail on the article explaining why.

 

[solarisguy]

@enemy85, thank you for the link!

I do not have time to digest the theory, yet. My first impressions from a quick read:

• many FreeNAS users use Advanced Format drives with 4k native sectors and seldom for large databases, while the experience described there comes predominately from 512-byte sector drives often used for databases;
• for a single GbE link probably other factors play a larger role, if the link cannot be saturated;
• it never occurred to me that there would be such a huge difference between AF and non-AF drives;
• compression argument might not be applicable to those users here that use FreeNAS primarily for audio, video and pictures.