ZFS performance bottlenecks: CPU or RAM or anything else?

[Steven Hartland]

My comment was targeted under the assumption of random IOPs workload, 
which is typically the case, where each RAIDZ group (vdev) will give 
approximately a single drive performance. For a pretty definitive guide 
/ answer see:
http://blog.delphix.com/matt/2014/06/06/zfs-stripe-width/

There's also some useful practical test results here:
https://calomel.org/zfs_raid_speed_capacity.html

On 18/05/2016 05:21, Alex Tutubalin wrote:
> On 5/18/2016 12:11 AM, Steven Hartland wrote:
>> Raidz is limited essential limited to a single drive performance per 
>> dev for read and write while mirror is single drive performance for 
>> write its number of drives for read. Don't forget mirror is not 
>> limited to two it can be three, four or more; so if you need more 
>> read throughput you can add drives to the mirror.
>
> Do I understand it correctly:
>
>  - single write of one large file (or singe local write to zvol shared 
> via iSCSI) will be local: single or only several metaslabs
>
>  - for RAIDZ each disk will get only part of throughput
>
>  - for mirror, each disk included in write will receive full data size 
> (and for single local write only limited number of disks to be 
> included in write)
>
> If so, raidz will have huge write performance benefit in my case: 
> single write of one large file.
>
> As for read speed, I hope to deal with it with large enough L2ARC on 
> SSDs.
>
>
>>
>> To increase raidz performance you need to add more vdevs. While this 
>> doesn't have to be double i.e. the same vdev config as the first it 
>> generally a good idea.
>
> Again, multiple vdevs will help for multiple parallel writes, but not 
> for single one?
>
> Alex Tutubalin
> _______________________________________________
> freebsd-fs at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-fs
> To unsubscribe, send any mail to "freebsd-fs-unsubscribe at freebsd.org"