Unsolved

3 Posts

3002

September 4th, 2020 11:00

Protection level +2d:1n or +4d:1n on 4 nodes

Hi,

I am considering about protection level of a new H500 with 4 nodes, 480TB.
How difference between protection level +2d:1n and +4d:1n?
Administration guide says [+2d:1n] is 6 + 2 (25%) and [+4d:1n] is 12 + 4 (25%).

Requested protection disk space usage
http://doc.isilon.com/onefs/9.0.0/help/en-us/GUID-CA694952-948F-4740-A174-E80D22516782.html

Does it mean parity overhead with +2d:1n and +4d:1n are as same as 25%?
And does same overhead mean both +2d:1n and +4d:1n have same 324 TiB usable space?
If so, +4d:1n seems to be better than +2d:1n because 2 more disk is okay to fail.


Regards,

YASU

Moderator

 • 

7.9K Posts

September 7th, 2020 08:00

Hello Yasu,

The protection percentage is the same for +2d:1n and +4d:1n.  The space should be the same as long as all the drives are the same size & speed.  

3 Posts

September 7th, 2020 17:00

Thank you for reply.
I find Isilon sizer shows performance of H500 (480GB) v8.2.2 as below;
Sequential Read
+2d:1n = 6.81 GBps 16 threads
+3d:1n = 6.47 GBps 16 threads
+4d:1n = 6.06 GBps 16 threads
So I will take +2d:1n.

Regards,

YASU

36 Posts

September 15th, 2020 09:00

There are two impacts to choosing a higher protection level:

  1. Performance. There is a small but measurable performance impact to choosing higher protection levels.
  2. Efficiency. In your case, the overhead on large files is the same, but the overhead on small files is larger because you need 4 FEC blocks regardless of whether you have 12 data blocks (12+4) or 1 data block (1+4).

In general, it is best to choose the recommended level. The system uses a "failure simulator" to generate statistical data and uses that data to determine the lowest protection level that guarantees an MTTDL of at least 5000 years. For Gen 6 nodes, that generally results in +2d:1n for smaller cluster/drive sizes and +3d+1n1d for larger clusters and drive sizes. Unless there are other compelling reasons/requirements to do so, using the recommended value is the right thing to do.

 

3 Posts

September 18th, 2020 04:00

Thank you for reply.
Okay I will take +2d:1n and check data 5000 years later.

Regards,

YASU

0 events found

No Events found!

Top