Write-Through and Write-Behind Caching Overview
Note
This feature is only available in NCache Enterprise Edition.
NCache supports write through caching which allows write operations to the back-end data source. In this way you can synchronize your cache and the master data source. In write-through caching, NCache updates cache store first and then applies that operation to the configured data source. For example, if a client application updates an entry in cache, then NCache will also update the configured data source (if write through is enabled).
Similarly, for write-through you need to implement WriteThruProvider Interface.
NCache framework will internally use this custom provider to perform write
operations on the back-end data source. Here you have to implement your custom
logic for write operations on the back-end data source. NCache will call your
provider behind write operations (Add, Insert, Remove/Delete) API calls with
write-thru. Currently NCache provides two modes for write-through caching.
- Write-through (Updates data source synchronously)
- Write-Behind (Updates data source asynchronously)
Note
NCache provides a performance counter for Write-Through operations per sec.
Write-Through
In Write-through caching, an operation is first applied on cache store and then synchronously updated to the configured data source. In write through mode, operations will be completed after NCache applies that operation on the back-end data source. You can use write-through caching if immediate database updates are critical and you need to update data source as soon as cache is updated.
Write-Behind
In Write-through, due to synchronous operations on data source, rate of operations on data source will be same as the rate of user operations on the cache. For applications with high user traffic, rate of user operation on cache can be very high which can resultantly overwhelm your data source. Also synchronous data source operations may affect response time of a user operation.
To overcome these problems, Write-behind can be used instead of write through. In write-behind, data source operations are performed asynchronously after NCache performs operations on cache store. After updating cache store, these operations are queued up and later applied to configured data sources asynchronously. Thus write-behind mode will enhance the response time of cache operations. NCache provides different configurations settings in write-behind to control operations flow on data source. For instance, you can specify the rate at which NCache will apply write behind operations on data source through Throttling.
Throttling
It indicates the number of operations applied on data source in a second. Default value for throttling is 500 ops/sec. You can change this value through backing source settings in NCache Web Manager.
Note
NCache provides a performance counter for Write-Behind operations performed per second.
Write-Behind Modes
NCache allows you to apply write behind operations individually or in batch. A write behind queue is maintained for write-behind operations. All write-behind operations will be queued in this queue and later applied to the data source according to the configured batched or non-batched mode. These two modes are explained below:
- Non-Batch Mode
By default, non-batch mode will be configured for write-behind operations. In this mode, operations in write behind queue will be applied one by one on data source according to the configured throttling rate. For example, if the throttling rate is 500 operations per second, NCache will apply write behind operations one at a time to data source and their applying rate will not exceed from 500 operations per second.
- Batch Mode
In Batch-mode, you can configure operation delay for write behind
operations, which indicates the time in milliseconds that each operation must
wait in the write-behind queue before applying it on the data source. By
default, its value is zero. In this mode, a batch/bulk of operations is selected
according to their operation delay. A dedicated thread periodically collects all
those operations which completed their delay interval after a configurable
interval called "batch-interval". Thus Batch-interval is the configurable
interval according to which NCache periodically checks for operation delay timed
out operations in write-behind queue. In short, a bulk of ready operations (that
completed their delay interval) are selected at every batch interval.
For example, if operation delay is configured as 1000ms and batch interval as 5 seconds, NCache checks the operations in write-behind queue after every 5 seconds (batch-interval) and selects all operations which have expired operation delays (all operations which are in queue for last 1000 milliseconds).
After a selection of operation bulk, these operations are then applied to the data source according to the configured throttling rate. Let's say a bulk of 1000 operations are selected from a write-behind queue, these operations are then applied to the data source in a batch of 500 operations (if throttling rate is 500 ops/sec) as the maximum operation applied to data source per second can't exceed to given throttling value.
You can specify an operation delay time ranging from seconds to days and months. In this way you can pause your operations on the data source by configurable amount of time. NCache also provides performance counters for write-behind queue, operations count and current batch operation count. Current batch operation count displays the number of operations selected in current batch interval for execution.
For write-behind, if batching is enabled, operations which are ready to be executed on data source are dequeued from write behind queue. The number of operations de-queued in the current batch interval will be displayed by the current batch operation count counter.
Write-Through Caching Operation Result
NCache provides you with the flexibility to synchronize write-through operations
in cache on the basis of its operation result. After applying an operation
(Add/Insert) on data source, you can specify operation status on the basis of
which NCache will synchronize the cache store. For example, in case of data
source operation failure, you can decide to remove that item from cache or to
keep it. You can also retry that operation on the data source. For this, you
have to specify Success/Failure/FailureRetry/FailureDontRemove as
DSOperationStatus of OperationResult. This is provided in both modes of
write-through caching i.e. write-thru/write-behind. Data source operation status
and their corresponding actions by NCache are described below:
Success: This means that the data source operation is successful and the item was added to the data source so NCache will keep it in the cache as well.Failure: This means that data source operation failed and the item could not be added to database, so NCache will remove it from the cache as well.FailureDontRemove: This means that the data source operation failed and the item could not be added to the database, but NCache will keep it in the cache.FailureRetry: This means that data source operation failed and the item could not be added to the database, so NCache will keep the item in cache and retry. Retries will be done as write behind operations.
Retrying Failed Operations
NCache allows you to retry operations in write-through/write-behind in case they are failed on data source. For this purpose, if you enable operation retrying and specify to retry a certain operation through provider, then NCache will retry that operation on data source. In case of write-through or write–behind, all retry operations will be re-queued to write behind queue, which means a write thru retry operation will be retried asynchronously as a write behind operation.
Note
NCache also provides a performance counter for Data source failed
operations/sec. Write operations performed on data source
returning Failure/FailureRetry/FailureDontRemove as DSOperationtatus of
OperationResult are counted per second by this counter.
NCache allows you to limit the number of failed operations to be retried. In
such a situation, you will mention the "Failed operation queue limit" through
NCache Web Manager, and if that limit exceeds you can evict failed operations
through a configurable eviction ratio. Here NCache will evict most retried
operations when retried queue is full. Each operation has associated the
RetryCount property which is incremented on each operation retried on the data
source.
For this, NCache provides a performance counter for the write-behind failure
retry count and write-behind eviction/sec. The Write-behind failure retry
counter will show the number of operations re-queued for retry. Data source
write operations returning FailureRetry as Status in OperationResult will be
requeued for retry. Whereas write-behind eviction/sec counter will display the
number of retry operations evicted per second.
Updating Cache after Data Source Operation
As stated earlier, in write-through caching, the operation is first performed on
cache store and then to data source. There may exist scenarios in which after
performing operation on data source, an item’s value becomes modified e.g. in
case of identity columns, its value will be modified by data source operation.
In such situation, data may become inconsistent in cache and data source. To
handle this, NCache allows you to specify whether to update data in cache after
data source operations or not. You can set flag "UpdateInNCache" to perform
operation (Add/Insert) again on cache store to make it synchronized with the
data source.
NCache updates operations in cache store, if specified, synchronously or asynchronously depending on the write-through caching mode. In write thru, synchronous updates will be applied and asynchronous in case of write-behind.
Note
NCache also provides a performance counter for data source updates/sec. It
displays the number of update operations per second in cache after
data source write operations. Data source write operations with UpdateINCache
flag set to true in OperationResult are then performed on cache. This
counter displays the number of these update operations performed on the
cache per second.
Hot Apply Support for Write-Behind Configuration
NCache supports hot apply for write-behind settings which allows you to change write behind configurations at runtime without stopping the cache. You can change almost all write behind configurable attribute through NCache Web Manager and NCache will incorporate those changes dynamically.
In this hot apply support, you can change the write-behind mode from batch to non-batch and vice versa. For instance, if you have changed the batch mode to non-batch, then NCache will ignore operation delay value and start executing operations individually. Also you can change the throttling rate at runtime according to your need. Similarly, operation delay, batch-interval, failed operation queue limit and eviction ratio can also change at runtime.
Warning
You can only change the "failed operation queue limit" in increasing fashion; otherwise NCache will use its default value for further operations.
Write Behind in Clustered Environment
As a write-behind queue is maintained for write-behind operations, a separate dedicated thread monitor executes its operation. The Topology level details for write-behind are mentioned below:
In the replicated cache topology, write behind queue will be maintained on all nodes, but write-behind async processor will be present on coordinator node only. It means that all write-behind operations will be performed through this node and replicated to other node queues cluster wide. In this way, if a node is down, than the next coordinator will perform all of the remaining write behind operations.
In the partitioned-replicated topology, write-behind queue is maintained on each active node and also replicated to its corresponding replicas. Each node will be responsible for its write-behind operation on data source.
In mirrored topology, write behind queue will be maintained on both active and passive nodes, but only the active node will be responsible to perform write-behind operations. Similarly, if the active node is down, then the passive will become active and perform the entire remaining write behind operations.
In partitioned topology, write-behind queue is maintained on each partition and every node will be responsible for its write-behind operations on data source.
In This Section
Write-Through Provider Configuration and Implementation
Explains the IWriteThruProvider interface and provides a sample implementation for the interface.
Using Write-Through with Cache Operations
Provides samples to use Write-Through with basic operations in NCache.
Using Write-Behind with Cache Operations
Provides samples to use Write-Behind with basic operations in NCache.
Monitor Write-Through Counters
Describes the performance counters provided by NCache to monitor Write-Through Caching.