FreeBSD iSER initiator

Jason Bacon jwbacon at tds.net
Thu May 7 14:03:30 UTC 2015


Many thanks to Mellanox for supporting FreeBSD!  Having the capabilities 
of FreeBSD available to Infiniband HPC clusters, etc. will be a great 
advantage for the entire industry...

On 5/7/15 6:30 AM, Sagi Grimberg wrote:
> Hello everyone,
>
> Mellanox Technologies is pleased to announce the availability of a new
> iSCSI Extensions for RDMA (iSER) initiator mode support for FreeBSD.
>
> What is it ?
>
> iSER is a computer network protocol that extends the iSCSI storage
> networking protocol to use Remote Direct Memory Access (RDMA). RDMA is
> provided by either RoCE (RDMA over converged Ethernet), iWARP or
> InfiniBand to enable efficient data movement using hardware offload
> capabilities.
>
> Why doing this ?
>
> RDMA infrastructure provides benefits such as Zero-Copy, CPU offload,
> Reliable transport, Fabric consolidation and many more. The iSER
> protocol eliminates some of the bottlenecks in the traditional
> iSCSI/TCP stack, provides low latency and high throughput and is well
> suited for latency aware workloads.
>
> What was added ?
>
> The iSER transport was added as another icl provider module, and it
> opens and RDMA channel to the target (icl_conn_connect). Other than
> that the addition of the iser module is pretty straight-forward.
>
> Some minor additions and changes were added to the iscsi layer for 
> iser support:
> - Not relying on io_received when handling a scsi response since
>   io_received is not maintained in iser (data is transferred via RDMA).
> - Added a PDU pointer to task_setup so iser can reference the scsiio in
>   its private pdu context.
> - Added a -T <transport> flag to user-space to indicate using iser.
> - Added iSER relevant information to the login request.
>
>
> The code is available at https://github.com/sagigrimberg/iser-freebsd
>
> The code maturity level is considered as alpha. It has gone through
> limited testing on top of 11-current using Mellanox ConnectX-3 adapters
> (Login, Logout, I/O). We tested against Linux LIO and TGT targets and
> we have several known issues in error flows and session recovery areas
> (Documented in release_notes.txt).
>
> We also did some initial performance measurements and we can see that
> iser achieves x20 BW (~5.4 GB/s) and x2 IOPs (~100 KIOPs) compared to
> iscsi/tcp for a single LUN device.
>
> We would like to get some feedback on the code in general, and
> specifically on some of the framework changes that we did in the iscsi
> common layers (and user-space). Moreover, feel free to try out the
> code, report BUGs (to Max or me at this point) and even provide fixes.
>
> Cheers,
>
> Max Gurtovoy & Sagi Grimberg
>
> Mellanox Technologies
> _______________________________________________
> freebsd-infiniband at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-infiniband
> To unsubscribe, send any mail to 
> "freebsd-infiniband-unsubscribe at freebsd.org"


-- 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   Jason W. Bacon
   jwbacon at tds.net

   If a problem can be solved,
   there's no need to worry.
   
   If it cannot be solved, then
   worrying will do no good.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~



More information about the freebsd-infiniband mailing list