[OmniOS-discuss] 2TB vs 4TB NVMe drives?

Stephan Budach stephan.budach at jvm.de
Tue Nov 14 15:54:04 UTC 2017


Hi Bob,

----- Ursprüngliche Mail -----
> Von: "Bob Friesenhahn" <bfriesen at simple.dallas.tx.us>
> An: "Stephan Budach" <stephan.budach at jvm.de>
> CC: "omnios-discuss" <omnios-discuss at lists.omniti.com>
> Gesendet: Dienstag, 14. November 2017 15:44:33
> Betreff: Re: [OmniOS-discuss] 2TB vs 4TB NVMe drives?
> 
> On Tue, 14 Nov 2017, Stephan Budach wrote:
> 
> > we are planning on purchasing a Supermicro NVMe server with 48 .U2
> > slots. I intended to initially load it with 24 x 2TB DC P4500,
> > leaving 24 slots empty.
> >
> >
> > Now… I've been also offered an Intel chassis with only 24 slots and
> > thus the offer also included the 4TB P4500s. Just without thinking
> > very long, I instinctively wayed towards the 2TB drives, mainly for
> > the reason, that should a drive really fail, I'd have of course a
> > longer resilver at hand, usind 4TB NVMe drives.
> >
> >
> > Which ones would you choose?
> 
> Assuming that OmniOS works with these devices at all, from a power
> consumption, heat, complexity, and reliability standpoint, the larger
> devices appear to be a win (1/2 the power and 2X the MTBF for the
> same
> storage capacity).  Resilver time is important but NVMe drives do not
> have the rotational latency and seek time issues of rotating media so
> resilver time should not be such an issue and there should only be a
> factor of 2 difference in resilver time.
> 
> A consideration is what zfs pool configuration you would be putting
> on
> these drives.  For throughput, more devices and more vdevs is better.
> It sounds like you would initially (and perhaps forever) have the
> same
> number of devices.
> 
> Are you planning to use zfs mirrors, or raidz2/raidz3?  What about
> dedicated zfs intent log devices?  If synchronous writes are
> important
> to you, dedicated zfs intent log devices should still help with pool
> performance and long-term health by deferring writes to the vdevs so
> writes can be larger and more sequential.
> 

Afaik, all the hardware is in the Illumos HCL, so this config should run fine under omniOS.
This setup is intended to replace my current ZFS-HA storage pools and it will be configured with zfs mirrors only, where the mirror vdevs will be built on iSCSI LUNs from "raw" devices, as much as you can get raw devices served by COMSTAR.

So, we will have these boxes serving each NVMe as a LUN to the RSF-1 nodes, which then will host 2 zpools of 6 mirror vdevs each. From thsoe zpools, the RSF-1 nodes will serve NFS to our Oracle VM cluster servers. 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 5546 bytes
Desc: not available
URL: <https://omniosce.org/ml-archive/attachments/20171114/d77bc359/attachment.bin>


More information about the OmniOS-discuss mailing list