Effect of Block Size on RAID Speed

[bwinter88]

Hi guys, I was finding conflicting information on how to determine the ideal block or chunk size when setting up a RAID so I did a little test myself. This is using two 1 TB SSD drives set up in a RAID 0 configuration using the NewerTech 6G-1e1i on a Mac Pro 5,1.

With the lowest setting block size of 16K, I got 658MBs write, and 906MBps read.

And with the highest block size of 256K, I got 687MBps write, and 910MBps read.

Now, I am a video editor, and the Blackmagic disk speed test is configured to test for speeds for a typical video editing scenario, so it's not representative of the performance you would get if you were, say, managing a database, but it pretty much confirmed my suspicions that with 2 drives, the block size really doesn't matter for big files since the load will be distributed pretty evenly across those drives. Hope this helps anyone.

 

[Gav2k]

Was always told it was 256 devided between the number of drives for smaller arrays. Larger it's 512 by the number of drives.

 

[AidenShaw]

Hi guys, I was finding conflicting information on how to determine the ideal block or chunk size when setting up a RAID so I did a little test myself. This is using two 1 TB SSD drives set up in a RAID 0 configuration using the NewerTech 6G-1e1i on a Mac Pro 5,1.

With the lowest setting block size of 16K, I got 658MBs write, and 906MBps read.


And with the highest block size of 256K, I got 687MBps write, and 910MBps read.


Now, I am a video editor, and the Blackmagic disk speed test is configured to test for speeds for a typical video editing scenario, so it's not representative of the performance you would get if you were, say, managing a database, but it pretty much confirmed my suspicions that with 2 drives, the block size really doesn't matter for big files since the load will be distributed pretty evenly across those drives. Hope this helps anyone.

A synthetic sequential benchmark like BlackMagic is next to useless for anything other than telling you how fast the BlackMagic benchmark runs on your system.

You need to test using the IO patterns important to your applications.

 

[bwinter88]

A synthetic sequential benchmark like BlackMagic is next to useless for anything other than telling you how fast the BlackMagic benchmark runs on your system.

You need to test using the IO patterns important to your applications.

Blackmagic test replicates the IO patterns important to my application—video editing. it's a good test.