The Open Systems Adapter (OSA): Delivering ~25 Years IBM Mainframe IP Connectivity

Recently in my day-to-day activities I encountered a 3172 controller and was reminded of my first such encounter, back in 1992.  This got me thinking; 25 years of IBM Mainframe IP connectivity!  The IBM 3172 Interconnect Controller allowed LAN-to-Mainframe interconnection and was the pioneering technology allowing IP data off-load activities.  Historically Mainframe data transfer operations, namely CCW I/O were dependant on a physical channel, where the 3172 was a stepping stone to the Open Systems Adapter (OSA) card in 1994, quickly superseded by the OSA-2 card in 1995.  From a performance viewpoint, the OSA/OSA-2 cards matched maximum ESCON speeds of 17 MB/S.

However, the introduction of the OSA-Express technology in 1999 dramatically increased throughput to ~ 333 MB/S.  The OSA-Express technology bypasses CCW channel-based I/O processing, connecting directly to the Self-Timed Inter-connect (STI) bus of Generation 6 (Retrofit to Generation 5) S/390 Mainframes.  Data is transferred directly to or from the high speed Mainframe memory OSA-Express adapter I/O port, across the STI bus, with no intervening components or processing to slow down the data interchange.  This bus-based I/O, a first for IBM Mainframe computing, significantly increases data transfer speeds, eliminating inefficiencies associated with intermediary components.

Additionally, IBM developed a totally new I/O scheme for the OSA-Express adapter. Queued Direct I/O (QDIO) is a highly optimized data queuing-based data interchange mechanism, leveraging from the message queuing expertise IBM acquired with their multi-platform MQSeries middleware solution.  The QDIO-specific S/390 hardware instruction for G5/G6 machines, delivered an application to-OSA signalling scheme capable of handling the high-volume, multimedia data transfer requirements of 21st Century web applications.  Where might we be without the 3172 Interconnect Controller and the MQSeries messaging solution?

Since OSA-Express2 the channel types supported have largely remain unchanged:

  • OSD: Queued Direct I/O (QDIO), a highly efficient data transfer architecture, dramatically improving IP data transfer speed and efficiency.
  • OSE: Non-QDIO, sets the OSA-Express card to function in non-QDIO mode, bypassing all of the advanced QDIO functions.
  • OSC: OSA-ICC, available with IBM Mainframes supporting GbE, eliminating the requirement for an external console controller, simplifying HMC and to the z/OS system console access, while introducing TN3270E connectivity.
  • OSN: OSA for NCP, Open Systems Adapter for NCP, eradicates 3745/3746 Front End Processor Network Control Program (NCP) running under IBM Communication Controller for Linux (CCL) requirements.  Superseded by:
  • OSM: (OSA-Express for zManager), provides Intranode Management Network (INMN) connectivity from System z to zManager functions.
  • OSX: (OSA-Express for zBX), provides connectivity and access control to the IntraEnsemble Data Network (IEDN) to the Unified Resource Manager (URM) function.

Returning to my original observation, it’s sometimes hard to reconcile finding a ~25 year old 3172 Controller in a Data Centre environment, preparing for a z14 upgrade!  In conjunction with the z14 announcement, OSA-Express6S promised an Ethernet technology refresh for use in the PCIe I/O drawer and continues to be supported by the 16 GBps PCIe Gen3 host bus.  The 1000BASE-T Ethernet feature supporting copper connectivity, in addition to 10 Gigabit Ethernet (10 GbE) and Gigabit Ethernet (GbE) for single-mode and multi-mode fibre optic environments.  The OSA-Express6S 1000BASE-T feature will be the last generation to support 100 Mbps link speed connections.  Future OSA-Express 1000BASE-T features will only support 1 Gbps link speed operation.

Of course, OSA-Express technology exposes the IBM Z Mainframe to the same security challenges as any other server node on the IP network, and as well as talking about Pervasive Encryption with this customer, we also talked about the increased security features of the OSA-Express6S adapter:

  • OSA-ICC Support for Secure Sockets Layer: when configured as an integrated console controller CHPID type (OSC) on the z14, supports the configuration and enablement of secure connections using the Transport Layer Security (TLS) protocol versions 1.0, 1.1 and 1.2. Server-side authentication is supported using either a self-signed certificate or a customer supplied certificate, which can be signed by a customer-specified certificate authority.  The certificates used must have an RSA key length of 2048 bits, and must be signed by using SHA-256.  This function support negotiates a cipher suite of AES-128 for the session key.
  • Virtual Local Area Network (VLAN): takes advantage of the Institute of Electrical and Electronics Engineers (IEEE) 802.q standard for virtual bridged LANs. VLANs allow easier administration of logical groups of stations that communicate as though they were on the same LAN.  In the virtualized environment of the IBM Z server, TCP/IP stacks can exist, potentially sharing OSA-Express features.  VLAN provides a greater degree of isolation by allowing contact with a server from only the set of stations that comprise the VLAN.
  • QDIO Data Connection Isolation: provides a mechanism for security regulatory compliance (E.g. HIPPA) for network isolation between the instances that share physical network connectivity, as per installation defined security zone boundaries. A mechanism to isolate a QDIO data connection on an OSA port, by forcing traffic to flow to the external network.  This feature safeguards that all communication flows only between an operating system and the external network.  This feature is provided with a granularity of implementation flexibility for both the z/VM and z/OS operating systems.

As always, the single-footprint capability of an IBM Z server must be considered. From a base architectural OSA design viewpoint, OSA supports 640 TCP/IP stacks or connections per dedicated CHPID, or 640 total stacks across multiple LPARs using a shared or spanned CHPID.  Obviously this allows the IBM Mainframe user to support more Linux images.  Of course, this is a very important consideration when considering the latest z13 and z14 servers for Distributed Systems workload consolidation.

In conclusion, never under estimate the value of the OSA-Express adapter in your organization and its role in transitioning the IBM Mainframe from a closed proprietary environment in the early 1990’s, to just another node on the IP network, from the mid-1990’s to the present day.  As per any other major technology for the IBM Z server, the OSA-Express adapter has evolved to provide the requisite capacity, performance, resilience and security attributes expected for an Enterprise Class workload.  Finally, let’s not lose sight of the technology commonality associated with OSA-Express and Crypto Express adapters; clearly, fundamental building blocks of Pervasive Encryption…

System z MLC Pricing Increases: Look After The Pennies…

Recently IBM announced ~4% price increases in z/OS Monthly License Charges (MLC) for selected Operating System and Middleware software programs and associated features. Specifically, price increases will apply to the VWLC, AWLC, EWLC, AEWLC, PSLC, FWLC and TWLC pricing metrics. Notably, SDSF price increases will be ~20% with Advanced Function Printing (AFP) product price increases of ~13-24%. In a global economy where inflation rates for The USA and Western Europe are close to 0%, one must draw one’s own conclusions accordingly. Lets’ not forget that product version changes typically have an associated price increase. From a contractual viewpoint, IBM only have to provide 90 days advance notice for such price changes, in this instance, IBM provided 150+ days advanced notice.

Price increases are inevitable and as always, it’s better to be proactive as opposed to reactive to such changes. As always, the old proverbs always make good sense and in this instance, “look after the Pennies and the Pounds will look after themselves”! This periodic IBM price increase is inevitable, but is not the underlying issue for controlling System z software costs. For many years, since 1994 to be precise, when IBM introduced Parallel Sysplex License Charges (PSLC), the need for IBM Mainframe users to minimize MSU usage has been of high if not critical importance. Nothing has changed in this 20+ year period and even though IBM might have introduced Sub-Capacity and specialty engines to minimize chargeable MSU usage, has each and every System z user optimized their MSU usage? Ideally this would not be a rhetorical question, rather being a “Golden Rule”, where despite organic CPU capacity increases of ~10% per annum, a System z environment could maintain near static IBM MLC software costs.

I have written several blog entries and presented on this subject matter over the years, for example:

The simple bottom line is that System z MLC software accounts for ~20-35% of the overall System z TCO, typically being the #1 expenditure item. For that reason alone, it’s incumbent for each and every System z user to safeguard they have the technical and commercial skills in place to manage this cost item, not as an afterthought, but inbuilt into each and every System z process, from application design, through to that often neglected afterthought, application tuning.

Many System z organizations might try to differentiate between a nuance of System and Application tuning, but such a “not my problem” type attitude is not acceptable and will be imposing a significant financial burden on each and every organization.

A dispassionate and pragmatic approach is required for optimizing System z CPU usage. In this timeframe, let’s examine the ~20% SDSF price increase. IBM will quite rightly state that in conjunction with their z/OS 2.2 release, there are significant SDSF product function advancements, including zIIP offload, REXX interoperability and increased information delivery. However are such function improvements over and above the norm and not expected as a Business As Usual (BAU) product improvements, which should be included in the Service & Support (S&S) or Monthly License Charges (MLC) paid for software?

In October 2013 I wrote a blog entry; Mainframe ISV Software: Is Continuous Product Improvement Always Evident? The underlying message was that an ISV should deliver the best product they can, for each and every release, without necessarily increasing software costs. In this particular instance, the product was an SDSF equivalent, namely (E)JES, which many years ago delivered all of the function incorporated in SDSF for z/OS 2.2, but for a fraction of the cost…

As of 1 November 2015, IBM will start billing cycles for Country Multiplex Pricing (CMP), which requires the October 2015 version of SCRT, namely V23R10. A Multiplex is defined as a collection of all System z servers in one country, measured as one System z server for software sub-capacity reporting. Sub-Capacity program utilization peaks across the Multiplex will be measured, as opposed to separate peaks by System z servers. CMP also provides the flexibility to move and run workloads anywhere with the elimination of Sysplex aggregation pricing rules.

Migrating to CMP is focussed on CPU capacity growth and flexibility going forward. Therefore System z users should not expect price reductions for their existing workloads upon CMP deployment. Indeed there are CMP deployment considerations. A CMP MSU baseline (base) needs to be established, where this MSU Base and associated MLC Base Factor is established for each sub-capacity MLC product and each applicable feature code. These MSU and MLC bases represent the previous 3 Month averages reported by SCRT before commencing CMP. Quite simply, to gain the most from CMP, the System z user must safeguard that their R4HA for each and every MLC product is optimized, before setting the CMP baseline, otherwise CMP related cost savings going forward are likely to be null.

From a very high-level management viewpoint, we must observe that IBM are a commercial organization, and although IBM provide mechanisms for controlling cost going forward, only the System z user can optimize System z MLC cost for their organization. Arguably with CMP, Soft-Capping isn’t a consideration, it’s mandatory.

Put very simply, each and every System z user can safeguard that they look after the Pennies (Cents) and the Pounds (Euros, Dollars) will look after themselves by paying careful attention to System z MLC software costs. Setting a baseline of System z MLC costs is mandatory, whether for the first time, or to set a new baseline for CMP deployment. Maintaining or lowering this System z MLC cost baseline should or arguably must be the objective going forward, even when considering 10% organic CPU growth, each and every year. System z decision-makers and managers must commit to such an objective and safeguard the provision of adequately skilled personnel to optimize such a considerable TCO cost line item (I.E. MLC @ ~20-35% of System z TCO). In an ecosystem with technical resources including DBA, Systems Programmer, Capacity Planner, Application Personnel, Performance Tuning, et al, why wouldn’t there be a specialist Software Cost Manager?

Let’s consider how even an inexperienced System z user can maintain a baseline of System z MLC costs, even with organic CPU capacity growth of 10% per annum:

  • System z Server Upgrade: Higher specification CPU chips or Technology Transition Offering (TTO) pricing metrics deliver 10%+ cost per MSU benefits.
  • System z Specialty Engines: Over time, more and more application workload can be offloaded to zIIP processors, with no sub-capacity MLC software charges.
  • System z Software Version Upgrades: Major subsystems such as CICS, DB2, IMS, MQSeries and WebSphere deliver opportunity to lower cost per MSU; safeguard such function exploitation.
  • Application Tuning: Whether SQL, COBOL, Java, et al, or the overall I/O subsystem, safeguard that latest programming techniques and I/O subsystem functions are exploited.
  • New Application Deployment: As and when possible, deploy new or convert existing workloads to benefit from the optimal MLC pricing metric; previously zNALC, nowadays zCAP.
  • Technical & Commercial Skills Currency: Safeguard personnel have the latest System z software pricing knowledge, ideally from an independent 3rd party such as Watson & Walker.

In conclusion, as householders we have the opportunity to optimize our cost expenditure, choosing and switching between various major cost items such as financial, utility and vehicle products. As System z users, we don’t have that option, only IBM provide System z servers and associated base architecture, namely the most expensive MLC software products, z/OS, CICS, DB2, IMS and WebSphere/MQ. However, just as we manage our domestic budgets, reducing power usage, optimizing vehicle TCO and getting more bang from our buck for financial products various, we can and must deliver this same due diligence for our System z MLC TCO. With industry averages of ~$500-$1000 per MSU for z/OS MLC software and associated annual expenditure measured in many millions, why wouldn’t any System z user look to deliver 10%+ cost per MSU optimization, year-on-year for their organization?

Clearly the cost of doing nothing in this instance, is significant, measured in magnitudes of millions, each and every year. Hence for System z MLC TCO optimization, looking after the Pennies is more than worthwhile, while the associated benefit of the Pounds, Euros or Dollars looking after themselves is arguably priceless.