Today's 1GBps Storage Vault Video

[Sunshine1868]

Can anybody explain why they chose RAID 6? Linus himself said that "the most important speed factor was writing to the drives". So why choose a RAID configuration that suffers from write speed slowdowns??

[Glenwing]

Can anybody explain why they chose RAID 6? Linus himself said that "the most important speed factor was writing to the drives". So why choose a RAID configuration that suffers from write speed slowdowns??

 

I presume because the second most important thing is redundancy

[Sunshine1868]

I presume because the second most important thing is redundancy

 

ok, so why not RAID 10.....? or a more massive spanning array?

[EvilLemur]

Well he made it clear that the SSD 2tb cache is in raid 10 while the HDD use the more redundant raid 6  and its the best of both worlds.

[Glenwing]

ok, so why not RAID 10.....? or a more massive spanning array?

 

RAID 6 provides better redundancy than RAID 10. Any 2 disks can fail without data loss, whereas RAID 10 might be able to lose 2 disks without data loss but it depends on which disks.

[Sunshine1868]

RAID 6 provides better redundancy than RAID 10. Any 2 disks can fail without data loss, whereas RAID 10 might be able to lose 2 disks without data loss but it depends on which disks.

I guess that's fair enough. still a write penalty that blew my mind a little

[leadeater]

Also FreeNAS itself does write caching even without the SSD tier, no parity RAID type should be used without write-back cache. Hardware RAID controllers should also have a BBU and upgraded cache when using RAID 5/6 else the performance will be terrible. Parity write performance issues were solved years ago before FreeNAS etc were a big thing.

[LinusTech]

I actually ended up using a dual parity configuration with unraid. Performance is pretty much single drive for both reads and writes from the HDD array, but when writing to the SSD cache (software RAID 10 until btrfs RAID5 gets more stable) it will be full 10gbit

[Sunshine1868]

I actually ended up using a dual parity configuration with unraid. Performance is pretty much single drive for both reads and writes from the HDD array, but when writing to the SSD cache (software RAID 10 until btrfs RAID5 gets more stable) it will be full 10gbit

1) so, just so I understand completely, basically what you're saying is you save to the SSDs (cache) and then they save to the HDDs (more permanent storage)... I like it.

 

2) HOLY SPITBALLS THE KING REPLIED TO MY POST #FanBoy

[Ramaddil]

While I agree the write penalty is significant, unless your operating on a raid card.  I am getting 300+mb writes with my card.  For me the little more write speed improvement is not worth the risk of losing my array.  Plus I don't want to give up half the amount of potential storage.. For me 24 (2TB) drives in raid 6 is roughly 44TB Raw, VS 24TB Raw. To me that is a waste of money.

[leadeater]

While I agree the write penalty is significant, unless your operating on a raid card.  I am getting 300+mb writes with my card.  For me the little more write speed improvement is not worth the risk of losing my array.  Plus I don't want to give up half the amount of potential storage.. For me 24 (2TB) drives in raid 6 is roughly 44TB Raw, VS 24TB Raw. To me that is a waste of money.

 

Yep, RAID 5/6 writes on true hardware RAID with BBU and write-back cache hasn't been a problem for 15 years. Misinformation seems to stick so badly yet correct information is forgotten in mere minutes

[handruin]

RAID 6 provides better redundancy than RAID 10. Any 2 disks can fail without data loss, whereas RAID 10 might be able to lose 2 disks without data loss but it depends on which disks.

 

This is one piece of the equation.  The other piece is that with more drives you'll also be more space-efficient using RAID 6 only using N-2.  RAID 10 you use 50% of your drive capacity for redundancy being N/2.

[handruin]

I actually ended up using a dual parity configuration with unraid. Performance is pretty much single drive for both reads and writes from the HDD array, but when writing to the SSD cache (software RAID 10 until btrfs RAID5 gets more stable) it will be full 10gbit

 

Why mess with btrfs when you could use a more-mature zfs?

[LinusTech]

Why mess with btrfs when you could use a more-mature zfs?

 

ZFS support on Linux not great and btrfs in active development right now - making big strides.

 

Something most people don't know is that FreeNAS is is FreeBSD based, NOT Linux based.

[leadeater]

Why mess with btrfs when you could use a more-mature zfs?

 

Watch the video when you can, he explains why. Basically it comes down to much better vendor support, don't want to spoil it for anyone. 

[handruin]

ZFS support on Linux not great and btrfs in active development right now - making big strides.

 

Something most people don't know is that FreeNAS is is FreeBSD based, NOT Linux based.

 

I would think it's the other way around.  ZFS is well-supported on Linux at this point in time; my company's enterprise storage product is based off it under CentOS and I've been using it for quite some time under Xubuntu.  I work with one of the developers/contributors of ZFS on Linux who is quite knowledgeable with ZFS.  Might be worth reconsidering it in the future.  :-)

[handruin]

Watch the video when you can, he explains why. Basically it comes down to much better vendor support, don't want to spoil it for anyone. 

 

I looked for the video before replying but don't see it?  Is this one on YouTube yet or Vessel?

[LinusTech]

I would think it's the other way around.  ZFS is well-supported on Linux at this point in time; my company's enterprise storage product is based off it under CentOS and I've been using it for quite some time under Xubuntu.  I work with one of the developers/contributors of ZFS on Linux who is quite knowledgeable with ZFS.  Might be worth reconsidering it in the future.  :-)

 

I thought there was a licensing issue or something? Was quite a while when I looked into it.

 

Iirc it's not (strictly speaking) the same ZFS depending on the platform and there are some customizations made for different solutions here and there.

[leadeater]

I looked for the video before replying but don't see it?  Is this one on YouTube yet or Vessel?

 

Vessel, hence the not wanting to spoil it.

[handruin]

I thought there was a licensing issue or something? Was quite a while when I looked into it.

 

Iirc it's not (strictly speaking) the same ZFS depending on the platform and there are some customizations made for different solutions here and there.

 

There is a licensing issue and that issue just keeps the two from being distributed together in the same package.  The Linux kernel is "GNU General Public License Version 2 (GPLv2)" and ZFS is "Common Development and Distribution License (CDDL)".  Both are free open source licenses but because of the differences in the licenses they cannot be included or distributed together.  There is more info here if you're curious.

 

You're right in that this port of ZFS (OpenZFS) is not the exact same as the original platform but it is feature-ready and comparable and the one to consider for Linux use.  ZFS on Linux is also production-ready in terms of strong data integrity, stability, and performance when configured properly.  Might make for a nice video in the future on the pros/cons of different file systems.  ;-)