What the documentation is trying to say is exactly what I described in (2), that the datanode is not involved in the data transfers but only registers metadata to the namenode. All data transfers happen between client and target directly. The text may be a bit confusing, we'll try to make that more clear.
-Patrick On Wed, Sep 25, 2019 at 7:46 PM Patrick Stuedi <pstu...@gmail.com> wrote: > Hi Shashank, > > Short answers to your questions: > > 1) The NVMf tier is a data tier, it stores data just like the RDMA tier or > the TCP tier. This is a bug in the documentation as far as I can see. > > 2) you need to start a NVMf datanode. The datanode will register the NVMf > resources and target with the Crail namenode, but it will not be involved > in data transfers. Data transfers take place between the client and the > target. You can use any target, the SPDK target or the kernel target. > Starting the target is outside of Crail. > > -Patrick > > > > > > On Wed, Sep 25, 2019 at 7:31 PM Gugnani, Shashank < > gugnan...@buckeyemail.osu.edu> wrote: > >> Hi, >> >> I'm trying to use the NVMf storage tier provided by Crail. >> However, I'm a little confused about how it works and the documentation >> is misleading and unclear. >> >> The documentation on the website ( >> https://incubator-crail.readthedocs.io/en/latest/config.html) says that >> the NVMf tier can only be used for metadata and not data, but the blog post >> ( >> https://crail.incubator.apache.org/blog/2017/08/crail-nvme-fabrics-v1.html) >> shows it being used to store data. So, which one is true? >> >> Secondly, with NVMf, do I need to run the datanode processes or does the >> NVMf target serve as the datanode process? >> Which NVMf target should I use? Can I use the SPDK NVMf target or does >> Crail provide its own version? >> If it does, then where do I find it and how can I get it running? >> >> >> Thanks, >> >> Shashank >> >
