SMC-R Protocol Strengthens Data Sharing Performance While Reducing CPU Costs

SMC-R provides efficient host-to-host direct memory communication without incurring significant TCP/IP processing overhead. It also protects organizations’ IT investments by allowing them to leverage existing TCP sockets-based applications and workloads while transparently exchanging data using RDMA over existing Ethernet infrastructures. SMC-R provides low-latency communications while preserving key qualities of service, such as network security, resiliency, scalability and load balancing.

To achieve this, SMC-R uses TCP to establish and terminate connections and monitor heartbeat functions. Figure 2 illustrates how this dynamic in-line negotiation begins by using TCP options to indicate whether both peers are SMC-R capable. If so, they exchange RDMA credentials within the TCP connection flows, similar to SSL negotiation. If SMC-R negotiations are successful, the TCP connection remains active but dynamically transitions to exchange data out-of-band, using RDMA in a manner transparent to applications. This model preserves the critical operational and network management features of TCP/IP.

The Benefits of TCP Preserved

The hybrid nature of SMC-R allows coexistence with TCP connection load-balancing solutions (e.g., Sysplex Distributor) typically deployed within z/OS Parallel Sysplex* environments for enhanced scalability, performance and high availability. When clients access z/OS servers via Sysplex Distributor, networking overhead and latency can be significantly reduced if VIPAROUTE is enabled and both client and server use SMC-R. That’s because TCP connection traffic will bypass the Sysplex Distributor node after the connection is established.

Many key TCP/IP security features—such as IP filters and System Authorization Facility-based network access control checks—are preserved with no administrative changes required when SMC-R is enabled. It’s fully compatible with TCP-based network encryption and authentication technologies such as Transport Layer Security and SSL. However, IP layer encryption/authentication technologies, such as IPSec, aren’t compatible with SMC-R and RoCE; SMC-R enablement is automatically bypassed for any connections that require IPSec.

SMC-R also preserves key qualities of service for network management, such as minimal IP configuration and operational changes, dynamic discovery of partner RDMA capabilities and dynamic RDMA connection setup.

Because RoCE traffic isn’t routable, communication solutions using RDMA require participating hosts to be on the same physical layer-2 network. For RoCE, this means each host must have direct access to the same physical Ethernet LAN—that is, reachable using a standard Ethernet switched fabric. A standard 10GbE Ethernet switch that’s 802.3x flow control-enabled can be used to attach to the RoCE Express feature. Converged Enhanced Ethernet capability isn’t required.

With z/OS, SMC-R visibility is provided through updates to traditional network management functions such as NetStat, Systems Management Facility, Resource Measurement Facility and Network Management Interface. The SMC-R architecture allows z/OS servers to exploit RDMA with minimal configuration and operational changes. Using SMC-R, RoCE Express doesn’t require any form of IP addressing, user-defined Media Access Control addresses or software stack IP interface definitions. RoCE Express is based on PCI Express architecture and defined in the System z I/O configuration, using Hardware Configuration Definition (HCD) and in the TCP/IP profile as PCI Function IDs (PFIDs).

PFIDs are logically and dynamically grouped with Open System Adapter-Express (OSA-Express) features, used for TCP connection establishment. Based on physical network association, this grouping is defined by a new network configuration option called Physical Network IDs (PNet IDs). These are defined for OSA and RoCE in HCD for each physical adapter port. System software and firmware dynamically learn the PNet ID associated with each adapter.

Gus Kassimis is a System z networking architect and designer in z/OS CommServer and one of the lead architects of the SMC-R protocol.

Jerry Stevens is a senior technical staff member with IBM Software Group and works in AIM Enterprise Networking Solutions Architecture Strategy and Design with a focus on communications hardware architecture. He has more than 25 years of experience with z/OS network communications.

comments powered by Disqus



2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.

Expanding Beyond Borders

IBM zEnterprise System announcement delivers a new dimension in computing.

Reducing Complexity

The next-generation System z server is more than just a fast, scalable solution

Address the IT Conundrum

A Smarter Computing-based infrastructure enables a smarter planet

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store
Mainframe News Sign Up Today! Past News Letters