In this e-Learning module, you will learn about key grid concepts, grid modes, copy modes, copy throttle, and host write throttle for the IBM(R) Virtualization Engine TS7700.
A TS7700 grid is two or more separate TS7700 clusters connected by means of a TCP/IP network. A grid can consist of TS7720 clusters, TS7740 clusters, TS7720 tape attach, or a combination of these. A grid can include a minimum of two clusters and a maximum of six clusters. A hybrid TS7700 grid consists of at least one cluster that is attached to a physical tape library and one cluster that is not.
In a grid, virtual volume attributes are synchronized across all clusters allowing the emulated virtual devices for each cluster to have access to all the volumes. During outages, through virtual volume replication, redundant copies are transparently made available through any other cluster. Even when a specific cluster does not contain a virtual volume copy, it can still access redundant copies contained within peer clusters using the TCP/IP infrastructure as a channel extender.
Let us look at an example. For a two-cluster grid, there are four operations.
Local write with no remote copy: includes writing the compressed host data to the local cache.
Local write with remote copy: includes writing the compressed host data to local cache and copying the data to the remote cluster in the grid.
Remote write with no local copy: includes writing compressed host data to the remote cluster in the grid.
Local read with a remote cache hit: the compressed host data is read from the remote cluster cache using the grid link.
A fifth operation might be available for balanced mode only. Data from the remote cluster is either a deferred or immediate copy. Compressed host data is received on the grid link, and copied into the local cache.
For more details, refer to the appropriate section in the Understanding, Monitoring and Tuning the TS7700 Performance white paper.
The TS7700 supports host connectivity in both a preferred or active-active balanced mode. Preferred mode is when a host or hosts are only connected to devices of one cluster. Active-active or balanced mode is when a host or hosts are simultaneously connected to two or more clusters. When in balanced mode, IBM z/OS(R) allocations are balanced across all clusters providing additional throughput and availability because all content which resides within a grid is accessible from any cluster. Scratch allocation assist provides a policy-managed method to direct allocations to one or more specific clusters when running in balanced mode.
In a multi-cluster grid, copies of volumes can be made between two or more clusters. The copies can have different priorities known as copy modes. There are four copy modes: synchronous, immediate, deferred, and time delayed. Synchronous mode simultaneously duplexes all host writes to two library locations, providing a zero recovery point objective for data written to virtual tape. Immediate copies, also referred to as rewind-unload copies, are completed prior to a volume rewind unload completion providing a guaranteed volume granular recovery point objective. Deferred copies occur asynchronously after the rewind-unload command completes. When immediate or synchronous copies cannot complete due to an outage or unavailable target, the immediate or synchronous copies will be managed as deferred copies. These immediate copies are referred to as immediate-deferred copies and synchronous copies are referred to as synchronous-deferred copies. Time delayed copies allows the user to set how long in hours they would like to defer the copy to the clusters selected. Users can also set the starting point of the delay to after volume creation or last accessed.
For more information, refer to the appropriate section in the Understanding, monitoring, and tuning the IBM System Storage TS7700 performance white paper.
Volumes in the source cluster are defined as either to remove from cache, PG0, or to keep in cache, PG1. The TS7740 and TS7720T copy PG0 volume data to back-end tape and then removes the data from cache as soon as possible. PG1 volume data is also copied to back-end tape, and then retained in the cache.
Inbound copy throttling can occur when the target cluster is a TS7740 or TS7720T and its disk cache has crossed certain residency boundaries such as too much content not yet pre-migrated to physical tape. Inbound copies of any type are then throttled similarly to new host I/O in order to reduce the amount of newly introduced resident content. The thresholds associated with this throttling can be tuned through the library request command. In addition, all TS7700 clusters can introduce an inbound copy throttle when other activities are falling behind, such as a large backlog of immediate copies.
The following are potential triggers of copy throttle: Immediate copy: immediate copies to other clusters, where this cluster is the source or target, are taking too long or are predicted to take too long. Cached copy backlog: TS7740 or TS7720T cache has crossed a boundary of too much content still needing to be replicated to one or more peer clusters. Pre-migrate: amount of data to be pre-migrated within a TS7740 or TS7720T is above threshold. Free space: invoked when a TS7740 or TS7720T cache is near full of any data and no minimal candidates for migration are present.
The deferred copy throttle (or DCT) is used to regulate outgoing deferred copies to other clusters in order to prefer host throughput. Adjusting the DCT value and threshold will allow you to tune the performance of the deferred copies when heavy host workloads occur. When the DCT threshold is reached due to increased host workload, the TS7700 adds a delay to each block of deferred copy data sent across the grid links from the cluster. The larger the configurable delay, the slower the overall grid replication performance becomes during the heavy host workloads. The DCT threshold can be changed using the host console request. This value is used to determine the average host I/O rate that would be best for maintaining deferred copy throttling. The default DCT threshold value is 100 megabytes per second of post-compressed data and the default DCT is 125ms per block. These throttling values can be disabled or reduced based on the throughput characteristics of the IBM Power7(R) based servers. For more information, refer to the appropriate section in the Understanding, monitoring, and tuning the IBM TS7700 performance white paper.
Immediate copy host write throttle can be turned on due to immediate copies taking too long between clusters in the grid. The throttling can occur at both the source and target of the immediate copy. The TS7700 will change an immediate copy to immediate-deferred if the immediate copy has not completed after 40 minutes. Because the immediate copy throttling is predictive, it can sometimes be too aggressive, in which case it should be disabled through the host console request command. If immediate copies begin to enter the immediate-deferred state, it should be re-enabled. For more information, refer to the appropriate section in the Understanding, monitoring, and tuning the IBM TS7700 performance white paper.
In this e-Learning module, you learned about grid concepts, grid modes, copy modes, copy throttle, and host write throttle. For more information about each of these topics, refer to the appropriate section in the Understanding, Monitoring, and Tuning the IBM TS7700 performance white paper, or see other topics in this knowledge center.