Overview of FDRPAS SWAPDUMP to a remote site via TCP/IP

Usually, FDRPAS is used to copy disks to new disks in the same data center. Both the source and the target devices are connected by channels to the CPU where FDRPAS is running. FDRPAS reads data from the source devices and writes it to the target devices.

FDRPAS also has the ability to copy to disks at a remote site where the disks are not connected to the CPU where FDRPAS is running. In this case, FDRPAS uses TCP/IP to send the data. The remote site must already have a running z/OS operating system. At the remote site, FDR/UPSTREAM for FDRPAS TCP/IP receives the data and writes it to the target disks. The initial z/OS system at the remote site does not have to have all of the installation's usual features, since it will soon be replaced by the z/OS system(s) that are being migrated; the initial z/OS system just has to be able to run TCP/IP and FDR/UPSTREAM for FDRPAS TCP/IP. This function of FDR/UPSTREAM is described in a separate FDRPAS for TCP/IP User Guide.

This operation can be used to move an entire data center to a new location. After the SWAPDUMP, the system(s) can be IPLed at the remote site.

This type of operation must always be a SWAPDUMP, not a SWAP, since it is not possible to SWAP to devices that are not connected to the LPAR where the SWAP is running.

As for any SWAP or SWAPDUMP, all of the LPARs where the source devices are online must have monitors running to detect updates. Since neither the main SWAPDUMP nor the monitors have access to the actual targets at the remote site, they instead use a “supertarget”. A supertarget can be any offline DASD device that is connected to the main SWAPDUMP LPAR and all of the monitor LPARs. One supertarget can be used by many SWAPDUMPs at the same time. The supertarget does not have to be as large as the volumes being processed, but it should be at least a 3390-9.

The monitors should be submitted to the internal reader by the SWAPDUMP jobs as just-in-time monitors, using a model jobstream in the PASJOB DD statement, as described in Overview of FDRPAS SWAPDUMP to a Remote Site via TCP/IP.

SWAPDUMP to a remote site supports source volumes with up to 400,000 cylinders.

Since the objective is to move the data center and all of its workload, it is expected that you will want all of the SWAPDUMPs to complete at the same time, so that the disks at the remote site are reasonably consistent. Many of the considerations in FDRPAS for Large Scale Synchronized Migration (Using CONFIRM) apply to this operation.

The SWAPDUMPs do not complete at exactly at the same instant. You will want to quiesce your LPARs as far as possible (so that no unnecessary jobs or online systems are running) during the split to get a consistent image of the data for use on the new LPARs.

It is likely that the SWAPDUMP will take many hours, or even several days, to copy and synchronize all of the volumes. One reason is that I/O over TCP/IP is slower than over a mainframe channel. Another reason is that the operation will probably include almost all of the volumes at the installation.

Simplified Job Setup

Traditionally, a large scale synchronized migration required many separate FDRPAS jobs with no more than 100 volumes each, using CONFIRMSPLIT=YES to tell FDRPAS to complete all of the SWAPDUMPs at the same time. For SWAPDUMP to a remote site, the setup has been simplified. A single SWAPDUMP job for a synchronized migration can now include several thousand volumes. With KEEPACTIVE=REPEAT, SWAPDUMP repeatedly cycles through all of the source volumes. MAXTASKS= specifies the number of volumes (up to 100) that are actively processed at the same time. As each volume is completed, a new volume is started, but the I/O intercepts are left in place for the “completed” volume, to continue to record updated tracks. At the end of a cycle of processing all of the volumes, SWAPDUMP pauses briefly, and then begins a new cycle. The first cycle may take a very long time, since it does the initial copy pass for each volume. Each later cycle only needs to recopy tracks that have been updated since the previous cycle, as recorded by the I/O intercepts, which have remained in place.

In case the first cycle takes too long (default of 24 hours), FDRPAS terminates the cycle after completing the volumes that are currently being processed, and then starts a new cycle. The objective is to avoid tying up too much common memory with track tables. On the next cycle, the volumes that were completed in the first cycle should be completed quickly, because they will only need to have updated tracks recopied. FDRPAS then proceeds to do the initial copy pass for the volumes that were not processed in the first cycle.

As a safeguard, there is a limit for how long a volume can go without being processed. If a volume has completed the initial copy pass, and 48 hours go by without a new pass being started, and the volume has tracks updated that require additional memory for the track table, then the track table is not extended, and the SWAPDUMP fails for that volume. The SWAPDUMP job terminates after the current or next cycle and has to be resubmitted. See Recovering From Errors in Remote SWAPDUMP.

The only cases in which you would need to have multiple SWAPDUMP jobs are if you have more than several thousand volumes to copy, or if you want to process more than 100 volumes concurrently. It is not likely that processing more than 100 volumes concurrently would be helpful, unless you have very high bandwidth on your TCP/IP connection.

Since there are typically only one or two jobs for a remote SWAPDUMP, the GENSWAP procedure (FDRPAS-GENSWAP) is not needed. However, depending on circumstances, it may still be beneficial to use GENSWAP.

If more than one SWAPDUMP job is run, we recommend running the SWAPDUMP jobs on the same LPAR, to make synchronization more convenient.

Completing the Operation

When you are ready to complete the SWAPDUMP, you should quiesce the activity on all of the LPARs as much as possible, such that no unnecessary jobs or online systems are running. Then, to tell FDRPAS to complete the operation, either issue a MODIFY jobname,KEEPACTIVE=DONE console command manually, or run a MONITOR TYPE=KEEPACTIVE job to issue the command automatically. A MONITOR TYPE=KEEPACTIVE job is preferable if there are multiple SWAPDUMP jobs, because MONITOR TYPE=KEEPACTIVE ensures that all of the volumes have completed the initial copy pass before it issues KEEPACTIVE=DONE. When KEEPACTIVE=DONE is issued, SWAPDUMP terminates the current cycle after the copy of the currently active volumes, performs one final cycle, and completes the operation for all of the volumes.

Unlike CONFIRMSPLIT=YES, where the SWAPDUMPS of all of the volumes are completed within seconds or a few minutes, the synchronization with KEEPACTIVE is less precise. The operation is completed by a full cycle through the volumes. The time for this cycle can be minimized by ensuring that as much activity as possible is quiesced, so that there are very few updated tracks to copy in the last cycle.

Any data that is written to DASD at the local data center after the SWAPDUMP operation is completed is not copied to the DASD at the remote data center. It is expected that when you run this type of migration for production, you are about to shut down the local data center.

  • The DASD at the local data center remains as a point-in-time backup as of the time the operation was completed.
  • You can run the migration as a test as many times as you want using as many volumes as you want, since it does not change the DASD at the local data center.

Volume States

During a remote SWAPDUMP, a volume is in one of four states:

Inactive:

The volume has not been processed yet, or processing has completed for this volume, or processing has failed for this volume.

Copy pass:

The volume is in the initial copy pass. FDRPAS is in the process of copying all used tracks on this volume to the target device at the remote site.

KEEPACTIVE:

The volume is between passes. The initial copy pass has been completed. One or more update passes may have been completed. The I/O intercepts remain in place for this volume, to continue to record updated tracks. An update pass will be needed to recopy any tracks that have been updated since the copy pass or the previous update pass.

Update pass:

FDRPAS is in the process of recopying updated tracks on this volume to the target device at the remote site.

A volume starts off as inactive, goes through the copy pass once, and then alternates between the KEEPACTIVE state and update passes. After the final update pass, the volume returns to the inactive state.

At any time, you can run FDRPAS with MONITOR TYPE=KEEPACTIVE to report the volume states, and possibly to take action depending on the volume states. The MONITOR TYPE=KEEPACTIVE job should be run on the same LPAR as the SWAPDUMP job.

  • MONITOR TYPE=KEEPACTIVE with PRINT=STATUS gives a detail report showing the state of each volume, produces a summary, and terminates. It is recommended to run this periodically during the initial copy cycle to monitor progress.
  • MONITOR TYPE=KEEPACTIVE without PRINT=STATUS periodically produces a one line summary showing the number of volumes in each state (inactive, copy pass, KEEPACTIVE, or update pass), and terminates when all of the volumes have completed the initial copy pass, i.e. are in either KEEPACTIVE state or an update pass. Before it terminates, it invokes FDREMCS. In this environment, FDREMCS reads its input from EMCSIN and writes its messages to EMCSPRT.
    • If EMCSIN is DD DUMMY, then the run has the effect of signaling when all of the volumes have completed the initial copy pass. This type of run is recommended toward the end of the initial copy cycle. When this job terminates, it means that all volumes have reached synchronization, and the operation is eligible to be completed.
    • EMCSIN can contain MODIFY KEEPACTIVE=DONE commands. Use this type of run to complete the SWAPDUMP operation.


 

Tip: For faster searching, add an asterisk to the end of your partial query. Example: cert*