Overview

A block diagram is a diagram of a system, in which the principal parts or functions are represented by blocks connected by lines that show the relationships of the blocks. To make an analogy to the map-making world, a block diagram is similar to a highway map of an entire nation. The major cities (functions) are listed, but the minor county roads and city streets are not. When troubleshooting, this high-level map is useful in narrowing down and isolating where a problem or fault is located.

 

In the same way, the following block diagram shows the principal parts and functions of a DXi system.

 

 

 

 

----------------------------------------------------------------------------------------

 

 

 

 

 

 

What's the Same Between DXi 1.x Software and DXi 2.x Software?

 

• The blockpool does replication. In addition, the blockpool and BPW are the only things that contain data.
• SNFS continues to be the file system for user data on DXi 2.0 software.

What's Different Between DXi 1.x Software and DXi 2.x Software?

• With DXi 2.x, QBFS doesn't exist, and the blockpool wrapper (BPW) is new. The BPW is the data pathy part, and the BPWAPI is a shared library. 
• All presentations (eg. CIFS) are modified to use shared memory. In addition, BPW feeds off of this shared memory to act upon the data.
• Ingest is truly “inline” and there's no longer an option for a Deduplication Policy choice. Data does not touch the disk until the final resting place. 

Impacts of Shared Memory (Pcache)
In the DXi 2.0 software architecture, the Backplane, or mid-layer, is a large shared memory, about 2.5 GB. On the small systems (DXi4500s and DXi6510s), it is 1 GB. The cache is divided up into pages 1 MB in size. All presentations, such as CIFS, are modified to use pcache shared memory. When writing data, the protocols take a page out of this shared memory and put the data there. BPW feeds off this shared memory to act on the data. Data does not touch the disk until the final resting place. BLOB tags now are handled by SNFS, not by QBFS.
 

Non-deduplicated data will follow this same data path through the BPW. It will not be written to the blockpool; it will be written to a non-dedup storage space instead.

 

There are several interfaces, generally referred to as ddupfs, which allow data movement into and out of the pcache. The cache is divided up into pages of 1 MB in size. This allows us to land data directly into RAM. The blockpool picks it up via shared memory interfaces, as well.

 

Data passes through the Backplane, which is an area of shared memory (called the Pcache). The back end of the Backplane feeds in to the Blockpool Wrapper (BPW). The BPW is the interface between protocols and the data stores (SNFS and BFST) and writes data to the blockpool (BFST).
 

All read operations, including tape creation, will be retrieved from the deduplicated content pool. There is no untruncated disk space.

 

Data comes in through a mid-layer referred to as the blockpool wrapper. The back end of the blockpool wrapper connects to the blockpool wrapper API. There are still a few interfaces directly talking to the blockpool (BFST server process) but the blockpool wrapper shields the blockpool for the most part. The blockpool wrapper is the interface between protocols and the data stores (SNFS and BFST). It is essentially an RPC server from the other side of that shared memory. This wrapper is really the glue that was QBFS and dedupd with DXi 1.x software.

  

The way the customer’s data is stored on disk in SNFS is through the blockpool. The metadata is in StorNext as raw files. The big change here, especially for people who are working inside the system, is that you used to be able to cd into the customer’s data and sniff around. It’s not there anymore. There’s no raw data available through a file system interface that you can just go and sniff. There are utilities available so that you can go and actually find the data.

---

What's Next?

DXi Software Architecture >

 

 

 

 

.