NFS/CIFS Benchmarking

[William Katsak]

What tools/packages are people using to do benchmarking on Linux? I've searched a little and haven't seen anything obvious other than NFSometer and SPEC (which is expensive).

 

[DrKK]

Benchmarking a FreeNAS install is very subtle. I am sure Cyberjock will come in here and give you a piece of his mind soon. :)

 

[William Katsak]

I am also interested in measuring the differences between RAIDZ1/2 with and without encryption. I know that just about any config can saturate a gigabit on an Avoton 2750, but I have 4 gigabit ports and want to see what my maximum is if I decide to put different loads on different ports.

Right now I am using some various block sizes, etc. with dd and writing various amounts of zero data (without compression enabled of course) and comparing, but I was wondering if there are any best practices or good tools for random read/write testing that people use.

 

[anodos]

My $0.02 is that you shouldn't bother benchmarking RAIDZ1 because you shouldn't be considering using it. Performance bottlenecks will depend heavily on your workload. Test with as realistic a workload as possible. This is because you may hit an IOPS bottleneck before CPU, in which case you may need more vdevs rather than turning off features like encryption.

 

[William Katsak]

I've seen mixed commentary regarding RAIDZ1/2. I understand the unaligned write business (with certain drive counts), and also the fact that statistically, when the number of drives goes up, the chance of a second failure on reslivering is high. However, I have a 4-drive FreeNAS Mini style system, and it seems that many people are successfully using RAIDZ1 in this scenario. 4 drives should be relatively ok in terms of likelihood of a second failure, no?

 

[anodos]

I've seen mixed commentary regarding RAIDZ1/2. I understand the unaligned write business (with certain drive counts), and also the fact that statistically, when the number of drives goes up, the chance of a second failure on reslivering is high. However, I have a 4-drive FreeNAS Mini style system, and it seems that many people are successfully using RAIDZ1 in this scenario. 4 drives should be relatively ok in terms of likelihood of a second failure, no?

http://en.wikipedia.org/wiki/Normalcy_bias
http://en.wikipedia.org/wiki/Neglect_of_prior_base_rates_effect

If your drives are above 1TB avoid RAID5 / RAIDZ1. For more information google it. I have personally seen the aftermath of multidrive failure in RAID5 (5 * 500GB drives) without adequate backups. The computer wasn't my responsibility, but it was not a fun experience that I want to repeat.