On Sat, 2018-03-17 at 08:48 +0000, Sitsofe Wheeler wrote:
> On 23 January 2018 at 17:50, Jens Axboe <axboe@xxxxxxxxx> wrote:
> > On 1/23/18 10:48 AM, Kip Ingram wrote:
> > > Hi Jens. I'm trying to do some NVME over IB testing, and it seemed
> > > appropriate to use the rdma engine in FIO. However, it's complaining
> > > in various ways:
> > >
> > > 1. It says I have to specify the device size. That's not a
> > > show-stopper, but it surprised me.
> > > 2. It says I have to use either read or write traffic - not mixed.
> > > That is a problem.
> > > 3. It says I can't use random addressing. That's also a problem.
> > >
> > > Colleagues have set up devices this way and used one of our internal
> > > tools to do mixed and random traffic.
> > >
> > > I don't see anything in the documents that imply mixed and random
> > > shouldn't work - is there something I'm overlooking?
> >
> > CC'ing the fio list, please direct questions there and not to me
> > directly.
>
> CC'ing some RDMA folks.
Hello Kip,
My understanding is that the purpose of the fio RDMA engine is to test RDMA
READ and WRITE and also that the fio RDMA engine cannot be used to measure
NVMeOF performance. If you want to test NVMeOF then you need to load the
NVMeOF initiator driver, tell that driver to connect to an NVMeOF target and
run fio against the block device created by the NVMeOF initiator driver.
Bart.
��.n��������+%������w��{.n�������^n�r������&��z�ޗ�zf���h���~����������_��+v���)ߣ�
