Forums

Home / Forums

You need to log in to create posts and topics. Login · Register

iSER support

Hello

Is iSER supported for iSCSI target?

Not at present. We do have it in our feature map, note that Ceph with rdma is still not a mature setup.

Hello
Are there any changes for this issue

Still Ceph rdma as well Ceph Seastor OSD engine is still some time to go. Implementing iSER now has no benefit.

Hello

If RDMA support appears in the Petasan for Ceph nodes replication, this will be great.
But initially I asked a question about iSER iscsi-traffic between the target and the initiator.

No we do not support it from target to initiator as there will not be any measurable benefit until Ceph supports RDMA and new Seastor OSD.

Hello.

First of all thank you for great product!

I am also asking for iSER / ROCe support on front lan for iSCSI connections.

ROCe cabable Mellanox and Broadcom swithces are wildly spead in enterprise.

iSER works over ROCe
RoCE = RDMA over Converged Ethernet
The RoCE v2 protocol exists on top of either the UDP/IPv4 or the UDP/IPv6 protocol.[2] The UDP destination port number 4791 has been reserved for RoCE v2.[10] Since RoCEv2 packets are routable the RoCE v2 protocol is sometimes called Routable RoCE[11] or RRoCE.[3] Although in general the delivery order of UDP packets is not guaranteed, the RoCEv2 specification requires that packets with the same UDP source port and the same destination address must not be reordered.[3] In addition, RoCEv2 defines a congestion control mechanism that uses the IP ECN bits for marking and CNP[12] frames for the acknowledgment notification.[13] Software support for RoCE v2 is still emerging. Mellanox OFED 2.3 or later has RoCE v2 support and also Linux Kernel v4.5.[14]
ROCE is about
  • Main pard implemented in NIC ASIC (primerelly Mellanox)
  • Etherenet swithes should support
       - PFC losless - Priority Flow Control
       - ECN - explicit congection notification
  • Etherenet swithes wih large buffers that do not drom packets

OpenStack Queens support iSER in The Block Storage service - cinder

"This driver supports different transport protocols to attach volumes, currently iSCSI and iSER.
volume_driver = cinder.volume.drivers.lvm.LVMVolumeDriver
iscsi_protocol = iser "
Something to read:

we will support iser once rdma is supported by ceph backend.

Quote from admin on February 4, 2021, 10:15 am

we will support iser once rdma is supported by ceph backend.

when I search the ceph source code, I find that src/msg/async/rdma has code for  rdma(iWarp type), does that mean rdma is now officially supported by ceph?

There was some older code for rdma support but not production grade. there is current effort to support rdma but the main issue is not just the transport layer  but the higher messaging, threading model, rocksdb speed, the Bluestore store engine itself, the current work will develop a new "Crimson" storage engine which addresses many of this. But this is not going to happen soon.

Current architecture delivers 0.3 ms read latency and 0.7-1 ms write latency, just moving to rdma will not provide noticeable improvement.