Cohesity: Architecture Concept and Terminology...

Learn Storage, Backup, Virtualization,  and Cloud. AWS, GCP & AZURE.

............................................................................................................................................................................
Whats up with Cohesity Architecture ?

• Uses Paxos algorithm for read consistency which is a mechanism to return read request from recently written value especially in a distributed filesystem.

• https://lamport.azurewebsites.net/pubs/paxos-simple.pdf

• Consistent Hashing to spread data across all nodes in a cluster.
• Data distribution using selected Erasure Coding (EC) or Replication Factor (RF) factor.

• Strict consistency : Non Disruptive Upgrades and operational function of non-disruptive service delivery at an event of disk or node failures i.e. strict consistency to support backup, restore, application data consistency and so on.

• SpanFS is an underlying web-scale file system which is a fully distributed filesystem which is where Cohesity Software Defined backup and recovery application “DataProtect” runs. SpanFS is what exposes NFS, SMB, and S3 Interfaces while it also manages the IO operation for all data written to or from the system.

• Distributed Lock Manager, manages concurrent access to the data repository and metadata

• Data Repository stores actual client data, such as network files, VMs, and databases ina. deduplication, compressed, and encrypted form.

• Metadata Store keeps track of all file data sitting across nodes,  Metadata store is based on Distributed Key-Value, that incorporates a fully redundant consistent, distributed NoSQL store for fast IO operations at scale.

• SnapTree is Cohesity’s builtin function that provides unlimited, frequent snapshots which  provides a distributed metadata structure  based on B+ tree concepts.

• Data Journaling: The SpanFS file system constantly looks at incoming requests and tries to estimate the IO pattern. Journal absorbs IOs and acts as write-cache which can be committed to disks later helping making data crash-consistent. It is part of the metadata and is replicated along with the File Metadata Store .

• Distributed Metadata Manager: On each node, the underlying SpanFS file system is used to write to disks. All file data is stored on the Distributed File Data Store. Distributed Metadata Manager maintains all metadata.

Pictorial Depiction Below:

You are Welcome :)

Source:https://info.cohesity.com/Cohesity-Fault-Tolerance-White-Paper.html