Economics and platform efficiency

Description: z/OS® Communications Server includes the following enhancements for economics and platform efficiency:
  • Shared Memory Communications over RDMA adapter (RoCE) virtualization - This function extends the Shared Memory Communications over Remote Direct Memory Access (SMC-R) function to TCP protocol workloads on the intraensemble data network (IEDN), and allow TCP/IP stacks on different LPARs within the same central processor complex (CPC) to share the same physical IBM® 10 GbE RoCE Express® feature.
    Restriction:
    • Connectivity to the intraensemble data network is allowed only when the central processor complex (CPC) is a member of an ensemble.
    • Each TCP/IP stack that shares the same physical 10 GbE RoCE Express feature must use a unique function ID (FID) and virtual function number (VFN) to represent the feature. Define the FID and VFN values in the Hardware Configuration Definition (HCD).
    Dependencies:
    • This function requires IBM z13® (z13) or later systems.
    • This function requires at least one IBM 10 GbE RoCE Express feature configured in the HCD with a FID and a VFN value.

    When change was introduced: z/OS V2R2.

  • SMC Applicability Tool (SMCAT) - The Shared Memory Communications (SMC) Applicability Tool (SMCAT) provides the ability to monitor and evaluate a TCP/IP network. You can use the evaluation to determine the applicability of SMC (over Remote Direct Memory Access (SMC-R) or using Direct Memory Access (SMC-D)) to your network environment. You do not need to enable the SMC function on any system, or enable any 10 GbE RoCE Express features, to use the SMC Applicability Tool.

    SMCAT can be used to monitor a TCP/IP stack for a set of configured destination IP addresses or subnets, and provide a report in the TCP/IP stack job log. The report provides details of the amount of TCP workload that can potentially use SMC if SMC is available.

    When change was introduced: z/OS V2R2.

  • VIPAROUTE fragmentation avoidance - A TCP/IP profile GLOBALCONFIG parameter, ADJUSTDVIPAMSS to adjust the TCP Maximum Segment Size (MSS) is added. Sysplex Distributor traffic that is routed by using VIPAROUTE adds a Generic Routing Encapsulation (GRE) header to the packet. Thus, the packet might be fragmented from the distributor to the target stack. The new function takes the GRE header into account when specifying the MSS value. It eliminates fragmentation by the distributor that would have been caused by the addition of the GRE header.

    When change was introduced: z/OS V2R2.

  • Enhanced Enterprise Extender scalability - The scalability of Enterprise Extender connections is improved. The overall performance is improved when there are large numbers of Enterprise Extender connections.

    When change was introduced: z/OS V2R2

  • Enhanced IKED Scalability - Internet Key Exchange daemon (IKED) scalability is improved. Performance is improved when large numbers of IKE peers attempt to negotiate IPSec SAs with z/OS IKED concurrently. The IkeSyslogLevel configuration parameter in iked.conf is updated and a thread identifier that precedes every IKED message written through syslogd is included.

    When change was introduced: z/OS V2R2

  • Increase single stack DVIPA limit to 4096 - The limit of total DVIPAs on a single stack is increased from 1024 to 4096. This allows more than 1024 application instance DVIPAs that are defined by VIPARANGE to be defined on a single TCP/IP stack.
    Restriction: The number of DVIPAs that are defined through VIPADEFINE and VIPABACKUP configuration statements is still limited to 1024.

    When change was introduced: z/OS V2R2

  • Shared Memory Communications over RDMA enhancements - A maximum transmission unit (MTU) value of 4096 is supported and autonomics performance for Shared Memory Communications over Remote Direct Memory Access (SMC-R) is enhanced.

    Previously, MTU values of 1024 and 2048 were supported. z/OS V2R2 Communications Server supports a new MTU value of 4096 for SMC-R.

    If you set the MTU size to 4096, you must also enable jumbo frames on all switches in the network path for all peer hosts.

    SMC-R autonomics provides the following performance enhancements:
    • SMCGLOBAL AUTOCACHE

      You can configure the TCP/IP stack to maintain statistics related to failed attempts to use SMC-R communications to specific destination IP addresses. When appropriate, the TCP/IP stack will direct future connections to those destination IP addresses to use TCP protocols instead of SMC-R, avoiding the overhead of unproductive attempts to establish an SMC-R link. This option is enabled by default.

    • SMCGLOBAL AUTOSMC

      You can configure the TCP/IP to analyze incoming TCP connections and dynamically determine whether SMC-R is beneficial to use for the connection. You can use this monitoring function to influence whether TCP connections to a particular server (port) use SMC-R, and to ensure that TCP connections use the most appropriate communications protocol (TCP or SMC-R). This option is enabled by default.

    When change was introduced: z/OS V2R2

  • TCP autonomic tuning enhancements - The following enhancements are made to automatically tune resources that are related to TCP connections. The enhancements are based on real-time data and can improve overall performance of TCP connections.
    • Dynamic Right-Sizing (DRS) and Outbound Right-Sizing (ORS) autonomics

      DRS and ORS are z/OS Communications Server optimizations that improve overall performance for certain high latency streaming workloads. z/OS Communications Server V2R2 lifts restrictions on which workloads and applications are eligible for each optimization and makes each optimization more sensitive to connection and CSM ECSA storage conditions. In addition, TCP/IP can stop and restart the DRS optimization dynamically for each connection based on the current system or application responsiveness.

    • Delayed transmission of acknowledgment autonomics

      z/OS Communications Server delays the transmission of acknowledgments on a TCP connection based on user configuration settings. z/OS Communications Server V2R2 provides autonomic capability to monitor the effectiveness of delaying the transmission of acknowledgments on a connection and a listener level. In addition, TCP/IP can stop and restart the delaying transmission of acknowledgments for each connection based on workload and application characteristics.

    • Fast Retransmit, Fast Recovery (FRR) autonomics

      z/OS Communications Server V2R2 extends the detection of out-of-order packets to lost packets during FRR recovery. When FRR recovery processing completes and no lost packets are detected , TCP/IP restores the transmission rates that were allowed before FRR recovery processing.

      When change was introduced: z/OS V2R2

  • 64-bit enablement of the TCP/IP stack, strategic DLCs, and CSM - The TCP/IP stack and the DLCs for OSA-Express in QDIO mode, HiperSockets, and RoCE-Express are enabled to fully use 64-bit virtual memory. These components run in AMODE64 and use virtual memory above the 2 GB bar, which significantly reduces the usage of data space, ECSA and private virtual storage below the 2 GB bar. The z/OS V2R2 TCP/IP stack 64-bit virtual memory support also improves networking scalability because TCP/IP's usage of data space, ECSA, and private virtual storage is not significantly affected by the scale of networking activity.

    In the following use cases, the z/OS V2R2 TCP/IP stack 64-bit virtual memory support might not provide the same level of performance as previous releases provide:

    • 31-bit network connectivity:

      Communications Server software that is related to OSA-Express in QDIO mode, HiperSockets, and RoCE-Express is updated to fully use 64-bit virtual memory. All other types of TCP/IP network connectivity, for example XCF, MPCPTP, LCS, or CTC, are 31-bit types and are updated to provide 64-bit stack compatibility. These drivers do not provide 64-bit exploitation. When you use the 31-bit types of network connectivity, your network performance and CPU cost might not be as efficient as it was in previous releases because extra data copies might be required.

      When 31-bit network connectivity is used for network traffic patterns going through z/OS, the impact might be more significant. Examples of these patterns include standard IP forwarding that is enabled by using IPCONFIG DATAGRAMFWD and sysplex distributor forwarding.

      Tip: Use VIPAROUTE over OSA-Express QDIO or HiperSockets for sysplex distributor forwarding to avoid using 31-bit network connectivity.

      The impact is based on the characteristics of your specific workloads such as message size and patterns, and environment. This is primarily an issue for streaming or bulk workloads.

    • Enterprise Extender (EE) support when 31-bit network connectivity is used:

      The z/OS Communications Server SNA (APPN) support for EE does not fully use 64-bit virtual memory. EE outbound processing is not affected. However, if inbound processing is connected to the network that uses 31-bit connectivity types, an extra copy of the inbound data might be required.

      Result: When you use OSA-Express in QDIO mode or HiperSockets for inbound EE processing, no additional data copies are required compared to V2R1.
      Tip: OSA-Express Inbound Workload Queueing (IWQ) support optimizes EE inbound traffic and further reduces the copies that are required.
    • Application socket APIs using CSM:

      Applications that use socket API semantics using Communications Storage Manager (CSM) managed memory can pass CSM data space or ECSA memory directly across the sockets API by using the UNIX System Services srx_np (BPX1SRX, BPX4SRX) callable services. The callable services continue to be supported without semantic changes. These services allowed data copies within the stack to be minimized, which reduces the stack send and receive processing cost.

      With the V2R2 64-bit virtual memory support, the performance benefits of the services are diminished. When you use the callable services, the TCP/IP stack copies the data between CSM buffers to 64-bit memory. The performance characteristics are similar to the sockets API with memory buffers provided by applications. In V2R2, you can continue to use applications that use the srx_np (BPX1SRX/BPX4SRX) services. Do not use these services for new applications.

      When change was introduced: z/OS V2R2

Reference information: See the following topics in z/OS Communications Server: New Function Summary for detailed descriptions that include any applicable restrictions, dependencies, and steps on using the functions:
  • Shared Memory Communications over RDMA adapter (RoCE) virtualization
  • SMC Applicability Tool (SMCAT)
  • VIPAROUTE fragmentation avoidance
  • Enhanced Enterprise Extender scalability
  • Enhanced IKED Scalability
  • Increase single stack DVIPA limit to 4096
  • Shared Memory Communications over RDMA enhancements
  • TCP autonomic tuning enhancements
  • 64-bit enablement of the TCP/IP stack, strategic DLCs, and CSM