Tiered Database StorageTiered Database Storage
This topic describes tiered database storage and provides recommendations for Hot, Warm, and Cold tier storage.
Starting with version 10.4, the Archiver service has the capability to be configured to use tiered storage. The concept of tiered storage is to put the most recent data on a Hot tier, which is the fastest storage available on the Archiver.
All services use the Hot tier by default.
The next tier is known as Warm and is typically cheaper and slower storage, such as a network-attached storage (NAS). The Warm tier contains older data; how old depends upon how much storage is allocated on the Hot tier and the average ingest rate. When the Hot tier reaches max utilization, the natural progression is to move the oldest data from the Hot tier to the Warm tier. When configured correctly, this happens automatically and is invisible to the end user. Queries and data access happen automatically no matter what tier (Hot or Warm) the data resides on. However, there can be a performance impact when accessing data on the Warm tier as compared to the Hot tier, because access times on the Warm tier are typically slower.
In addition to Hot and Warm, there is also a Cold tier. The Cold tier is only used as a staging area for offline backup. NetWitness Core services do not access data on the Cold tier. NetWitness Core services move the oldest data to the Cold tier and consider it abandoned (the service no longer accesses the data). This data can then be backed up to long-term storage like tape for possible restoration months or even years later, depending on requirements. The backing up and subsequent removal of data on the Cold tier must be handled outside of NetWitness Core services via scripts or other processes.
If the Cold tier becomes full because external processes are not removing data in a timely manner, this causes the NetWitness Core service to eventually stop the ingestion of new data until the problem is corrected.
When moving data to the Cold tier, NetWitness recommends that the directory remain on the same mount point as where it is being moved from. Therefore, if the files are coming from the Warm tier, it is far better for performance reasons to set the Cold tier directory on the same file system. The reason for this is that the service attempts to simply move the file and directory to the Cold tier, which is a nearly instantaneous operation on the same file system. If the move fails, the fallback is to copy the data to the Cold tier, which takes more processing time and causes additional I/O contention on the tier from which it is being copied.
The tiers of storage capabilities are used by the Archiver. You can configure Archiver to only use Hot storage (the default), Hot and Warm, or all three (Hot, Warm and Cold). All services must use Hot, you cannot configure a service to only use Warm. Data flows from Hot to Warm and finally to Cold. You can also skip Warm and go from Hot to Cold. If Cold (offline) storage is not configured, the oldest data is deleted on the last configured tier, which has been the standard operating procedure.
The typical Archiver deployment sets all the databases to unlimited size (packet.dir, meta.dir, session.dir, index.dir, and optionally the Warm tier variants), which means that the size specifier is left off or set to zero. This lets the databases and index grow unbounded. Instead of each database managing their own size and rolling out only when each individual database exceeds their configured size, Archiver rolls out everything together using the
command. This enables the databases and index to roll out in unison. For more information on sizeRoll, see "Asynchronous Rollover" in
Archiver is typically configured to place the index, session, meta, and packet (log) DB on the same volume, instead of multiple volumes like a Concentrator or Decoder. Although this can potentially cause more I/O contention when concurrent reads happen across multiple databases, it also maximizes overall retention. Because all databases are on the same volume, they are configured to roll out together, which minimizes orphaning of data. Decoder and Concentrator are configured for maximum I/O speed, but can suffer from estimates on the proper volume sizing.
For example, if the session DB is too large, it may have enough storage for six months of retention, whereas the meta DB and index only have retention for four months. Because the session, meta DB, and index are intricately tied together, the shortest retention period for all three define the overall retention period (in this case, four months). Retention of individual databases is mostly affected by factors beyond our control, such as traffic captured, meta generated (parsers, feeds, rules) and filtering. The databases are easily resized by a simple configuration change, but this usually also involves changes at the hardware and file system level to adjust partitions, which complicates dynamic resizing. Archiver avoids these problems by using a single volume for everything, with the trade-off of somewhat slower I/O speed.