From mboxrd@z Thu Jan  1 00:00:00 1970
Content-Type: multipart/mixed; boundary="===============4927666149135570897=="
MIME-Version: 1.0
From: Walker, Benjamin <benjamin.walker at intel.com>
Subject: Re: [SPDK] Nvmf target namespace specification
Date: Fri, 04 May 2018 18:09:30 +0000
Message-ID: <1525457369.46960.7.camel@intel.com>
In-Reply-To: CANvN+ekNNiKhh9x09dRL6CwMHM-v-e9heKo5ExZ0qBQ8_ZEN2w@mail.gmail.com
List-ID: <spdk@lists.01.org>
To: spdk@lists.01.org

--===============4927666149135570897==
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable

On Fri, 2018-05-04 at 17:56 +0000, Andrey Kuzmin wrote:
> Thanks for the quick turnaround, Daniel.
> =

> On Fri, May 4, 2018, 20:48 Verkamp, Daniel <daniel.verkamp(a)intel.com> w=
rote:
> > Hi Andrey,
> > =

> > Do you mean on the host (initiator) side, bdev/nvme?  When you are
> > connecting to a NVMe-oF target with the NVMe bdev (either via the [Nvme]
> > configuration file section or the construct_nvme_bdev RPC method), you =
are
> > specifying the address to a subsystem, and all namespaces in that subsy=
stem
> > will be exposed as bdevs.  There is currently no way to specify just a
> > single namespace within the subsystem, but we could probably add this if
> > there is a reasonable use case that requires it.
> =

> I don't think it as strong as required, but the case that has triggered my
> question above was trying to run bdevperf with its target device specific=
ation
> against a specific namespace at thee nvmf target. =

> =

> As is (with effectively just the controller specified), it actually runs
> prerry well, but I assume it does so against all namespaces the target ex=
ports
> via the specified subsystem.

When you connect to a controller with the NVMe-oF host, it will create a
separate bdev for each namespace automatically. If you do something like th=
is:

[Nvme]
  TransportID "trtype:RDMA <stuff>" Nvme0

Then it will create one bdev per namespace with the names Nvme0n1, Nvme0n2,=
 etc.

However, when you go to run bdevperf, there isn't actually a way to say to =
just
run against one of those namespaces. That's probably a reasonable addition =
to
the tool.

(Also, we should consider promoting the tool from a test tool in test/bdev =
to an
example in examples/bdev.)

> =

> Regards,
> Andrey
> > If you are talking about the target side, you can configure namespaces =
to
> > have a particular ID using the optional second parameter to the Namespa=
ce
> > line in the [SubsystemN] config sections, or via the optional namespace=
.nsid
> > parameter to the nvmf_subsystem_add_ns RPC method.
> > =

> > Thanks,
> > -- Daniel
> > =

> > > -----Original Message-----
> > > From: SPDK [mailto:spdk-bounces(a)lists.01.org] On Behalf Of Andrey K=
uzmin
> > > Sent: Friday, May 4, 2018 10:35 AM
> > > To: Storage Performance Development Kit <spdk(a)lists.01.org>
> > > Subject: [SPDK] Nvmf target namespace specification
> > > =

> > > Spdk nvmf docs suggest nvmf targets to be specified via
> > > traddr/addrfam/traddr/trsvcid tuple. Is there any way to specify the
> > > target controller namespace as well?
> > > =

> > > Regards,
> > > Andrey
> > > _______________________________________________
> > > SPDK mailing list
> > > SPDK(a)lists.01.org
> > > https://lists.01.org/mailman/listinfo/spdk
> > _______________________________________________
> > SPDK mailing list
> > SPDK(a)lists.01.org
> > https://lists.01.org/mailman/listinfo/spdk
> =

> -- =

> Regards,
> Andrey
> _______________________________________________
> SPDK mailing list
> SPDK(a)lists.01.org
> https://lists.01.org/mailman/listinfo/spdk

--===============4927666149135570897==--