Hardware raid or ZFS?
- Thread starter Alessandro 123
- Start date
RAID10 is too risky in a 8 disks server.
You could loose up to 4 disks at once, but only 1 for each mirror pair. With 8 disks, there is a very high probability that 2 disks fails from the same mirror.
RAID-6, in my case, is not an option.
Did you do any performance comparison between hardware raid and zfs raid ?
You could loose up to 4 disks at once, but only 1 for each mirror pair. With 8 disks, there is a very high probability that 2 disks fails from the same mirror.
RAID-6, in my case, is not an option.
Did you do any performance comparison between hardware raid and zfs raid ?
I can also evaluate RAIDZ-3 (this force me to use ZFS as I don't have any hardware raid capable of doing RAID7)
But I don't know if ZFS is slower than hardware raid
But I don't know if ZFS is slower than hardware raid
Hi,
I made an test between zfs raid-z2 and areca hw raid-6... zfs lost on my testsystem (Amd 8350 + 32GB Ram). A huge different between both solutions (app. 15% iowait between).
Udo
Hi,
I assume it's depens on your system. I guess with an fast powerfull system you can beat an slow raidcontroller... But if you use an fast one you should get better values with hw-raid (but you are more flexible with zfs).
My testsystem uses 6 sata-disks first with zfs and after high IO-Wait (and other ugly things, like no expansion of an zfs raid-z2) I use an areca raid-controller and the IOwait drops a lot.
Udo
I think ZFS is mutch more better then HW raid. You can search the forum about that. But yes... ZFS need more RAM and zfs need very good HW. So only enterprisedisks, more more more RAM. But the features are worth. But yes it's your decision. HW Raid is easier to handle, and the servers are cheaper, then real ZFS ready serves.
When you have time you can build your own zfs hw. Supermicro, 96GB RAM, WD Red Pro and Enterprise SSD from Intel.
When you have time you can build your own zfs hw. Supermicro, 96GB RAM, WD Red Pro and Enterprise SSD from Intel.
I only use enterprise disks (SAS 15K) and our new servers has at least 64GB RAM
But i don't know the ZFS performance. I think I have to do some tests.
But i don't know the ZFS performance. I think I have to do some tests.
Yes to that. It's up to you
Do not be unsettled by the high I / O. This will distort at HW Raid.
Hi,
forgot that I use an cache SSD too for zfs: Intel DC S3700 200GB... and enterprise-disk with sata has no realy an performance benefit more an reliability benefit (I don't talk from 5400rpm disks).
We will see the tests from Alessandro.
Udo
Keep this in mind if you decide to deploy hardware raid. Make sure that it has adequate NVRAM.
https://blogs.oracle.com/bonwick/entry/raid_z
Personally, I think a ZFS pool of mirrored pairs would work quite nicely.
https://blogs.oracle.com/bonwick/entry/raid_z
Personally, I think a ZFS pool of mirrored pairs would work quite nicely.
https://www.howtoforge.com/how-to-benchmark-your-system-cpu-file-io-mysql-with-sysbench
Sysbench is a good choice, please share your benchmark numbers when you have done the testing!
Whether you use HW RAID or LVM, replication groups (eg RAID 10) are almost always better performing for VM workload then erasure coding (eg parity RAID.) Regardless, given enough RAM ZFS is superior to HW RAID not because it performs better (although that can be the case) but because of LVM-filesystem integration in ZFS. Simply put, ZFS is smarter then HWRAID.
Exactly, but in my case, I have to do RAID6/RAIDZ2 and thus I have to evaluate performance carefully
I dont understand what you are saying. If you begin with the premise that you have to do RAID6, what is the purpose of evaluating performance? do you have minimum performance criteria defined?
If I may make a suggestion, you might want to begin with defining your performance and fault tolerance criteria first and decide on technique as a consequence, not the other way around.