[SPDK] Re: PMEM over RDMA (RPMEM) Support Status in SPDK

[SPDK] Re: PMEM over RDMA (RPMEM) Support Status in SPDK

[Harris, James R]

[-- Attachment #1: Type: text/plain, Size: 2584 bytes --]

Hi Joe,

There are a few ways to access persistent memory over RDMA:

1) The bdev module that Paul referenced - see module/bdev/pmem.  This case uses PMDK to create a libpmemblk pool, creates a bdev from it, which can then be exposed over RDMA using the NVMe-oF target
2) PMDK's librpmem - this provides a mechanism to remotely replicate a PMDK memory region over the RDMA protocol - you can find more details on pmem.io
3) Provision a devdax namespace and register it for RDMA access directly (note: you'll still need something on the target side to ensure data is properly flushed to the persistence domain)

If you're interested in #2 or #3, you may want to check out the pmem mailing list on Google Groups.

-Jim


On 3/2/20, 3:44 PM, "Luse, Paul E" <paul.e.luse(a)intel.com> wrote:

    Hi Joe,
    
    I can give you a little on PMEM in general, will let someone else jump in with any NVMeoF stuff... Today SPDK uses PMDK to support PMEM in two major ways:
    
    - we can expose a pmem region as a bdev through the block device module for pmem
    - our compression bdev uses pmem behind the scenes to track metadata for compressed data stored on other media
    
    Thx
    Paul
    
    
    On 3/2/20, 3:28 PM, "Gruher, Joseph R" <joseph.r.gruher(a)intel.com> wrote:
    
        I'm just curious about the support for PMEM over RDMA Ethernet in SPDK.  I assume it will be similar functionally to NVMe/NVMeoF, but for PMEM... eventually.
        
        I was wondering if you can point me to a website, whitepaper, or person where I could get a general sense of SPDK support for PMEM, both implemented as of today, and future roadmap/schedule.
        
        I'd love to test it out in my lab when it is mature enough for that to be a reasonable thing to attempt, as I am fairly well equipped with RDMA capable NICs and with PMEM equipped systems, and experienced with using SPDK with NVMe/NVMeoF, but I'd like to wait until it is reasonably mature, so trying to get a sense of where we are today, and when might be a reasonable target date to try and engage with some basic testing / proof-of-concept type work.
        
        Thanks,
        Joe Gruher
        
        _______________________________________________
        SPDK mailing list -- spdk(a)lists.01.org
        To unsubscribe send an email to spdk-leave(a)lists.01.org
        
    
    _______________________________________________
    SPDK mailing list -- spdk(a)lists.01.org
    To unsubscribe send an email to spdk-leave(a)lists.01.org
    

[SPDK] Re: PMEM over RDMA (RPMEM) Support Status in SPDK

[Luse, Paul E]

[-- Attachment #1: Type: text/plain, Size: 1503 bytes --]

Hi Joe,

I can give you a little on PMEM in general, will let someone else jump in with any NVMeoF stuff... Today SPDK uses PMDK to support PMEM in two major ways:

- we can expose a pmem region as a bdev through the block device module for pmem
- our compression bdev uses pmem behind the scenes to track metadata for compressed data stored on other media

Thx
Paul


On 3/2/20, 3:28 PM, "Gruher, Joseph R" <joseph.r.gruher(a)intel.com> wrote:

    I'm just curious about the support for PMEM over RDMA Ethernet in SPDK.  I assume it will be similar functionally to NVMe/NVMeoF, but for PMEM... eventually.
    
    I was wondering if you can point me to a website, whitepaper, or person where I could get a general sense of SPDK support for PMEM, both implemented as of today, and future roadmap/schedule.
    
    I'd love to test it out in my lab when it is mature enough for that to be a reasonable thing to attempt, as I am fairly well equipped with RDMA capable NICs and with PMEM equipped systems, and experienced with using SPDK with NVMe/NVMeoF, but I'd like to wait until it is reasonably mature, so trying to get a sense of where we are today, and when might be a reasonable target date to try and engage with some basic testing / proof-of-concept type work.
    
    Thanks,
    Joe Gruher
    
    _______________________________________________
    SPDK mailing list -- spdk(a)lists.01.org
    To unsubscribe send an email to spdk-leave(a)lists.01.org