Multinode Persistent Sessions: The Sysplex Data Sharing End Game

1
Multinode Persistent Sessions The Sysplex Data
Sharing End Game
Bryant L. Osborn Bank of America
2
Introduction

• APPN- (Advanced Peer-to-Peer Networking) VTAM
  facility introduced in the late 1980s.
• HPR - (High Performance Routing) extension to
  reroute sessions around links that have failed.
• What does this have to do with the future of data
  sharing? Everything.

3
Definitions

• Scalability - a measure of the practical limits
  to how large a system can be.
• CEC - Central Electronics Complex, or mainframe,
  box, footprint, etc.
• CPC - Central Processor Complex, or a CMOS CEC.
• CS/390 - New name for VTAM

4
Sysplex Benefits
(Besides software discounts)

• Improved scalability
• Continuous availability

5
Improved Scalability
Irwin F. Kraus, The Parallel Sysplex as SMP
Viewing Performance, Capacity, and Scalability
through a Familiar Lens, CMG96 Proceedings,
December, 1996
6
Improved Scalability

• For MVS, maintaining the general case has too
  much overhead to go much above 12 processors.
• Sysplex no longer supports the general case.
• Sysplex builds systems by adding MVS images to a
  Sysplex

7
Improved Scalability
When a user architects a parallel sysplex, the
user decides how much data is to be shared,
usually much less than 100. The user decides how
many processing nodes will be used and what
capacity and performance. The user-system-archite
ct has a lot of control over minimizing overhead
and maximizing scalability in a parallel sysplex
computer - Irwin Kraus
8
Improved Scalability

• MVS goes massively parallel. Parallel Sysplex
  increases MVS scalability to make truly gigantic
  workloads possible.
• But wait! Hardware improvements have resulted in
  CECs with 1000 MIPS
• Which way to go?

9
Continuous Availability

• As workloads get larger and larger, the risk from
  an outage becomes greater and greater
• Parallel Sysplex may allow continuous
  availability for overnight batch workloads, but
  what about interactive workloads with logged-on
  users?

10
Continuous Availability
The 64,000,000 question What good is it to have
a single gigantic workload if the organization
cannot afford even a brief outage of the entire
system? Is anyone really willing to put all
their proverbial eggs in one basket, and run the
risk that a single outage could take down the
entire workload?
11
Single-Node Persistent Sessions

• A VTAM facility
• There are two kinds of persistent sessions. The
  kind that exists today is called single-node
  persistent sessions.
• VTAM keeps session state information about VTAM
  users. When a subsystem crashes, VTAM can
  maintain the sessions for a specified period of
  time.

12
Single-Node Persistent Sessions
13
The Benefits of SNPS

• Saves users from having to re-establish VTAM
  sessions (VTAM keeps ACBs open)
• Saves application time to clean up failed
  sessions, but
• Users can do no work until the application is
  brought back up, and
• What happens if VTAM itself fails??

14
What if . . .
Wouldnt it be nice is there was a way that users
could be reconnected to another MYCICS on CEC-B
by reconnecting them through a VTAM-B? What a
great idea! But is it possible? What you call it?
15
Multinode Persistent Sessions

• Users of a failed application are reconnected
  through a different VTAM to another application
  on another LPAR. How is this possible?
• Uses APPNs High Performance Routing (HPR)
  facility
• Allows fast recovery from failures of all kinds

16
Multinode Persistent Sessions

• SNPS stored user session information in
  dataspace. Where is the only place this
  information could be stored in a Sysplex where
• a) all VTAMs can get to it, and
• b) information will not get trashed by any
  failure?
• Requires a coupling facility and more that one
  LPAR

17
Multinode Persistent Sessions

• CICS implementation - MVS Automatic Restart
  Manager will start another region on another
  LPAR. VTAM will use HPR to perform a path switch
  to move the connection endpoint
• IMS implementation - will probably allow
  reconnection to an existing IMS. Have users
  open multiple ACBs at logon time, and switch
  between them???

18
VTAM Primer

• Node - is the endpoint of a communication link
• Subarea Networks - hierarchical networks
• APPN Networks - Peer-oriented networks

19
Essential Subarea Networks

• Key word is hierarchy
• VTAM is king of the hierarchy. Also called a
  System Services Control Point (SSCP).
• Network Control Program (NCP) in a communications
  controller is second
• Everything else is a peripheral node that comes
  in third

20
Essential Subarea Networks

• LU (logical unit) - is a port where users connect
  to the node. Every LU is owned by a VTAM (SSCP).
  LUs request request network services from VTAM
  (SSCPs).
• PU (physical unit) - manage links and routing.
  (Similar to APPN control points.)

21
Subarea Network Nodes

• Type 5 Node - VTAM (SSCP)
• Type 4 Node - NCP in a communications controller.
  Requires a type 5 node to provide all network
  services
• Type 2 Node - Peripheral node that cannot perform
  routing. They are dependent on the VTAM (SSCP).
• Type 2.1 Node - Peripheral node with limited
  peer-to-peer capabilities. Can be independent of
  the VTAM (SSCP) if using LU 6.2 communication.

22
Subarea Network Nodes
23
Subarea Networks

• All paths must be manually defined. Route
  selection is chosen from from predefined routes.
• Virtual Routes (VRs) - logical paths on which
  sessions are carried. SSCP (type 5) and NCP
  (type 4) nodes establish and maintain VRs.
• Class of Service (COS) - sessions are assigned to
  VRs on the basis of an assigned class of service.
• Management and Control Sessions - SSCP-SSCP,
  SSCP-PU, and SSCP-LU

24
Essential APPN Networks

• Key word is peer which means of equal
  standing
• Network Nodes (NNs) - provide all network
  services
• End Nodes (ENs) - provide LU ports and rely on
  Network Nodes (NNs) for the rest.

25
Essential APPN Networks

• Control Point (CP) - every APPN node has a
  control point from which LUs request network
  services. An APPN control point is similar to
  subarea PU.
• Transmission Group (TG) - links between APPN
  nodes.

26
APPN Network Nodes
27
APPN Networks

• APPN networks are self-defining. NNs maintain a
  distributed directory containing the
  characteristics and current status of each node
  and link.
• Central Directory Servers (CDSs) - Large networks
  have one or more CDS to hold the distributed
  directory.
• RTP Pipes - HPR groups sessions with the same
  Class of Service (COS) together into logical
  connections called RTP pipes. RTP pipes can be
  switched to a different path without affecting
  the sessions it carries.

28
APPN Networks

• Management and Control Sessions - CP-CP sessions
  between adjacent nodes
• Interchange Node (ICN) - special kind of APPN NN
  used to connect APPN and subarea networks.
  Supports hybrid VR-TG (virtual route based
  transmission group) communication and data flows.
• Dependent LU Server (DLUS) - special kind of APPN
  NN used to provide SSCP services to dependent LUs
  in a subarea network.

29
MNPS Example
30
MNPS Example

• The endpoint of an interchange node (ICN) can be
  either a transmission group (TG) or a virtual
  route (VR). Since it could be either, an
  RTP-capable path cannot be guaranteed, so IMS3 is
  not eligible for MNPS.
• If an LU were owned by EndNode1 (connected
  directly), the session with IMS1 is not
  recoverable because there is no HPR path to
  switch.

31
The Requirements for MNPS

• All VTAMs with MNPS applications (CICS, IMS, DB2)
  must be defined as APPN end nodes supporting RTP
• All VTAMs with MNPS applications must be
  connected to a MVS coupling facility that
  contains the VTAM structures
• MVS/ESA V5R2 or later
• Coupling Facility Control Code (CFCC) Level 1 or
  later
• All VTAMs that own MNPS applications must be V4R4
  or later

32
Conclusions about MNPS

• MNPS Fulfills the Sysplex Promise. When the
  risks become trivial, the size of data centers
  and workloads will explode.
• Data Sharing ? DTR ? Persistent Sessions
• MNPS Will Cut Across Organizational Lines.
  Implementation may be difficult because support
  will cut across support groups.

33
Questions???