Re: ZFS implementation reasons and challenges in OSv

[Dor Laor]

On Thu, Oct 5, 2017 at 11:29 AM, Kaeyan Jones  wrote:

> Hi Dor,
>
> I very much agree that ZFS is the best filesystem one could adopt, huge
> fan of it myself. Given the dependency on RAM for storage performance in
> ZFS are there any adverse side effects to deployments of OSv for bursty
> workloads, particularly I/O latency sensitive ones?
>

Our ZFS snapshot from FreeBSD can use some updates and isn't as performing
as XFS on some cases.


>
> Is OSv using the OpenZFS implementation illumos? I've been assuming it is
> since in the USENIX ATC 2014 paper ZFS's selection was from FreeBSD and
> FreeBSD borrows theirs from illumos.
>

At the time we needed there wasn't a separate repository and later we were
too busy but consuming openZFS directly is the best choice today.


>
> Was BRTFS considered? Is the GPL licensing a deal breaker?
>

BTRFS has its caveats and BGP is a no go for a unikernel that resides
within the same addr space as the app.


>
> Is hardware passing not necessary then? From my understanding of the
> illumos implementation of ZFS (in using FreeNAS), ZFS requires direct
> hardware control; virtualization requires hardware passing of disks and the
> use of RAID controllers requires JBoD. To do otherwise and not give ZFS
> direct hardware control of the disk is an invitation to data loss. Is this
> not similarly the case for ZFS in OSv?
>
> Given the performance, features and stability of ZFS coupled with OSv's
> improved network stack, is OSv particularly popular for usage in
> network/storage workloads?
>

It's a less common case for unikernels.


>
> Thank you,
> Kaeyan
>
>
> On Oct 5, 2017 11:26 AM, "Dor Laor"  wrote:
>
> Hi Kaeyan,
>
> ZFS was and still is the best filesystem with a BSD license we could adopt.
> There is no problem with hardware control that we know about. Moreover,
> hypervisors
> do virtualize the hardware so you can have similar features in a VM (like
> multiqueue and NVMe)
> and even direct assignment.
>
> Best,
> Dor
>
> On Thu, Oct 5, 2017 at 9:16 AM, Kaeyan Jones  wrote:
>
>> Hello,
>>
>> What were the reasons and rationale behind using ZFS for the main file
>> system of OSv?
>>
>> As I understand ZFS, it isn't a file system that plays nicely with
>> virtualization and demands hardware control. Did this not prove a problem
>> for OSv? How extensive is the testing of the ZFS implementation?
>>
>> Thank you,
>> Kaeyan Jones
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "OSv Development" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to osv-dev+unsubscr...@googlegroups.com.
>> For more options, visit https://groups.google.com/d/optout.
>>
>
>
>

-- 
You received this message because you are subscribed to the Google Groups "OSv 
Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to osv-dev+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: ZFS implementation reasons and challenges in OSv

[Kaeyan Jones]

Hi Dor,

I very much agree that ZFS is the best filesystem one could adopt, huge fan
of it myself. Given the dependency on RAM for storage performance in ZFS
are there any adverse side effects to deployments of OSv for bursty
workloads, particularly I/O latency sensitive ones?

Is OSv using the OpenZFS implementation illumos? I've been assuming it is
since in the USENIX ATC 2014 paper ZFS's selection was from FreeBSD and
FreeBSD borrows theirs from illumos.

Was BRTFS considered? Is the GPL licensing a deal breaker?

Is hardware passing not necessary then? From my understanding of the
illumos implementation of ZFS (in using FreeNAS), ZFS requires direct
hardware control; virtualization requires hardware passing of disks and the
use of RAID controllers requires JBoD. To do otherwise and not give ZFS
direct hardware control of the disk is an invitation to data loss. Is this
not similarly the case for ZFS in OSv?

Given the performance, features and stability of ZFS coupled with OSv's
improved network stack, is OSv particularly popular for usage in
network/storage workloads?

Thank you,
Kaeyan

On Oct 5, 2017 11:26 AM, "Dor Laor"  wrote:

Hi Kaeyan,

ZFS was and still is the best filesystem with a BSD license we could adopt.
There is no problem with hardware control that we know about. Moreover,
hypervisors
do virtualize the hardware so you can have similar features in a VM (like
multiqueue and NVMe)
and even direct assignment.

Best,
Dor

On Thu, Oct 5, 2017 at 9:16 AM, Kaeyan Jones  wrote:

> Hello,
>
> What were the reasons and rationale behind using ZFS for the main file
> system of OSv?
>
> As I understand ZFS, it isn't a file system that plays nicely with
> virtualization and demands hardware control. Did this not prove a problem
> for OSv? How extensive is the testing of the ZFS implementation?
>
> Thank you,
> Kaeyan Jones
>
> --
> You received this message because you are subscribed to the Google Groups
> "OSv Development" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to osv-dev+unsubscr...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
>

-- 
You received this message because you are subscribed to the Google Groups "OSv 
Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to osv-dev+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: ZFS implementation reasons and challenges in OSv

[Dor Laor]

Hi Kaeyan,

ZFS was and still is the best filesystem with a BSD license we could adopt.
There is no problem with hardware control that we know about. Moreover,
hypervisors
do virtualize the hardware so you can have similar features in a VM (like
multiqueue and NVMe)
and even direct assignment.

Best,
Dor

On Thu, Oct 5, 2017 at 9:16 AM, Kaeyan Jones  wrote:

> Hello,
>
> What were the reasons and rationale behind using ZFS for the main file
> system of OSv?
>
> As I understand ZFS, it isn't a file system that plays nicely with
> virtualization and demands hardware control. Did this not prove a problem
> for OSv? How extensive is the testing of the ZFS implementation?
>
> Thank you,
> Kaeyan Jones
>
> --
> You received this message because you are subscribed to the Google Groups
> "OSv Development" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to osv-dev+unsubscr...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
>

-- 
You received this message because you are subscribed to the Google Groups "OSv 
Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to osv-dev+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.