Super Fast Local Workloads with LXD, ZFS, and Juju

#1

Post by mitechie » Fri, Feb 12, 2016, 2:47 PM UTC

Super Fast Local Workloads with LXD, ZFS, and Juju
jorgecastro.org

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#2

Post by matt_wulfeck » Fri, Feb 12, 2016, 3:54 PM UTC

LXD is great. It lowers significantly the mental burden of running something in containers because they behave essentially like a VM. That being said LXD and juju are very Ubuntu things. LXD needs wide adoption for it to be attractive as a serious workload.

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#3

Post by hoistor » Fri, Feb 12, 2016, 3:59 PM UTC

I don't think the author of this post really understands ZFS.

Is that a 4 disk mirror? You'd be much faster using a raid10 there.

ZFS 'cache' devices are't for write cache either, you'd want a log device for write perf reasons if that's your constraint... Even then it's unlikely to make a difference on a workstation workload with 16GB of memory..

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#4

Post by marcoceppi » Fri, Feb 12, 2016, 4:00 PM UTC

matt_wulfeck wrote: post #2

LXD is great. It lowers significantly the mental burden of running something in containers because they behave essentially like a VM. That being said LXD and juju are very Ubuntu things. LXD needs wide adoption for it to be attractive as a serious workload.

There are LXD images for Ubuntu, CentOS, Gentoo, Plamo, and other Linux operating systems. So you can really boot most any Linux as you would a VM. Between that and it's use as a hypervisor in OpenStack it gets flexed pretty heavily. Though I agree, more people should be using it!

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#5

Post by justinsaccount » Fri, Feb 12, 2016, 4:16 PM UTC

hoistor wrote: post #3

I don't think the author of this post really understands ZFS. Is that a 4 disk mirror? You'd be much faster using a raid10 there. ZFS 'cache' devices are't for write cache either, you'd want a log device for write perf reasons if that's your constraint... Even then it's unlikely to make a difference on a workstation workload with 16GB of memory..

Yeah.. it is a subtle difference between

  zpool create home mirror a b c d cache e

and

  zpool create home mirror a b mirror c d cache e

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#6

Post by markramm » Fri, Feb 12, 2016, 4:17 PM UTC

This is a great development experience for distributed software. I'm really looking forward to Ubuntu 16.10 -- a lot of things seem to be coming together there.

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#7

Post by dr_zoidberg » Fri, Feb 12, 2016, 4:34 PM UTC

justinsaccount wrote: post #5

hoistor wrote: post #3

I don't think the author of this post really understands ZFS. Is that a 4 disk mirror? You'd be much faster using a raid10 there. ZFS 'cache' devices are't for write cache either, you'd want a log device for write perf reasons if that's your constraint... Even then it's unlikely to make a difference on a workstation workload with 16GB of memory..

Yeah.. it is a subtle difference between zpool create home mirror a b c d cache e and zpool create home mirror a b mirror c d cache e

Could you elaborate on the semantics of "cache"? I do understand the mirror keyword, but I know next to nothing about ZFS and don't really get what "cache" means, considering bouth your and OPs comment.

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#8

Post by ochoseis » Fri, Feb 12, 2016, 4:38 PM UTC

hoistor wrote: post #3

I don't think the author of this post really understands ZFS. Is that a 4 disk mirror? You'd be much faster using a raid10 there. ZFS 'cache' devices are't for write cache either, you'd want a log device for write perf reasons if that's your constraint... Even then it's unlikely to make a difference on a workstation workload with 16GB of memory..

I think he might just be glossing over some details? At any rate, from what I understand ZFS has a 'zil' write cache for small log writes, and a l2arc for common reads that don't fit into memory. This is just a dev setup, but for real work I think you'd want your zil to also be mirrored so you don't lose writes if an ssd goes bad. For anyone interested this link has a few more details: http://www.45drives.com/wiki/index.php/FreeNAS_-_What_is_ZIL...

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#9

Post by sparkiegeek » Fri, Feb 12, 2016, 4:44 PM UTC

markramm wrote: post #6

This is a great development experience for distributed software. I'm really looking forward to Ubuntu 16.10 -- a lot of things seem to be coming together there.

Oh? I'm sure 16.10 will be great but there's a bunch of things coming in 16.04 which have me excited!

Re: Super Fast Local Workloads with LXD, ZFS, and Juju

#10

Post by justinsaccount » Fri, Feb 12, 2016, 4:54 PM UTC

dr_zoidberg wrote: post #7

justinsaccount wrote: post #5

Earlier quoted context omitted.

Yeah.. it is a subtle difference between zpool create home mirror a b c d cache e and zpool create home mirror a b mirror c d cache e

Could you elaborate on the semantics of "cache"? I do understand the mirror keyword, but I know next to nothing about ZFS and don't really get what "cache" means, considering bouth your and OPs comment.

You can use a fast disk like an SSD as a 2nd level cache device after ram. So reads will first check the main ram cache first, then the ssd cache disk before finally hitting the underlying storage.