Authors

Mike Ham Technical Director of Applications Engineering, Storage Accelerators

Pinaki Chanda Director of Software Engineering

Introduction

The MaxLinear’s Panther III storage accelerator (Panther III) has been developed to optimize the performance and efficiency of storage appliances by offloading complex tasks that require significant CPU cycles to execute. The key features of Panther III are as follows:

• Data reduction
  • High-quality compression to minimize the size of data before storage.
  • Hash generation to drive deduplication to further reduce the size of data before storage.
• Data integrity—Panther III provides the most robust protection against data corruption available.
  • Real-time verification (RTV)—All encode operations are automatically followed by a decode operation and verification that the fully decoded data matches the original data.
  • Note: Transform commands are not completed until all requested data protection/verification operations are completed successfully.
  • Panther III provides full support for the T10-DIF, T10-DIX, and NVMe Protection Information (PI) data protection functions.
• Data security—Panther III supports NIST-certified Advanced Encryption Standard (AES) security standards used for SSL and IPsec transforms. They are optimal for securing data at rest (on the storage device), being transferred to/from the remote client or to/from a backup/replication storage appliance.
• Highest performance—Support for 200Gbps throughput with the lowest latencies available.

This document focuses on how Panther III provides the highest level of data integrity available.

Data Integrity

What is data integrity and why is it important?

Data integrity is the ability to ensure that data being transformed is not corrupted during the process. This includes protections for all data residing on the device, data moving through the device, and data being actively transformed. Without robust data integrity capabilities, data to be transformed could be silently corrupted (undetected) and stored. The corruption may go undetected until the data is decoded, by which time it may be too late to recover it since the original data is no longer available.

By implementing industry-leading data integrity features, Panther III can protect against silent data corruption for both encode and decode operations. Should data corruption occur, it can be detected by the Panther III device and reported. Panther III has the most robust data integrity capabilities of any solution on the market.

Panther III’s data integrity features are as follows:

• Real-time verification
• Internal data path protection
• Memory protection
• PCIe packet protection
• NVMe PI data protection fields
• Robust compression algorithm support
• PCIe Advanced Error Reporting (AER)

Real-Time Verification (RTV)

Panther III supports the use of RTV, which MaxLinear highly recommends. When enabled, all encode operations are automatically and transparently fully decoded. The fully decoded data is then verified against the original data to ensure that all transform operations have been completed successfully. Only when this transform verification is complete is the command completion notification sent. In this way, Panther III can self-detect any errors and report them if they occur in real time, thus avoiding any risk of silent data corruption, with no impact on performance and minimal overhead due to Panther III's heavily pipelined architecture.

 

The RTV function for hash operations is implemented as two parallel engines whose two outputs are verified byte by byte since hash operations have no decode capability.

One potential cause of an error on a Panther III device would be an alpha particle hit resulting in memory or logic disruption. Panther III goes beyond detecting and reporting every error occurrence. When an error is detected and reported by Panther III, the driver automatically reprocesses the command using the algorithm software library that implements all the transforms supported by the device. If successful, the command is completed with a notification that an error has been reported by the Panther III device and that the command has been successfully processed by the software. If unsuccessful using the software, the command is completed reporting a failure due to a command submission error. This helps identify the cause of the failure, while facilitating the integration of Panther III during the development process.

T10-DIF, T10-DIX, and NVMe PI Data Protection

Panther III supports the following industry-standard data protection features:

• T10-Data Integrity Field (T10-DIF)
• T10-Data Integrity List (T10-DIX)
• NVMe Protection Information (NVMe PI)

These industry-standard data protection features extend data integrity beyond the boundary of the Panther III storage accelerator and beyond the processor root complex (RC). They are all derived from the Small Computer System Interface (SCSI) T10-DIF standard managed by the T10 Technical Committee.

The T10-DIF and T10-DIX fields are 8-byte fields that contain three sets of information:

• The 2-byte Application (App) tag, which contains customer-defined information.
• The 2-byte Reference (Ref) tag, which is also customer-defined, but typically represents a storage logical block address (LBA) and increments with each T10-DIF/T10-DIX field.
• The 4-byte Guard block, which contains a CRC generated on the preceding data that the DIF/DIX field protects.

                                                                                                                        
 

Each T10-DIF/T10-DIX field is generated for a data block defined as a sector and protects this data block. Each sector is traditionally 512 bytes but can be any power of 2 from 512 bytes upwards (that is, 512 bytes, 1024 bytes, 2048 bytes, and so on). The only difference between the T10-DIF and T10-DIX fields is that the T10-DIF fields are inserted directly into the data at the end of each data sector they protect, whereas the T10-DIX fields are generated and managed as a separate list from the data.

NVMe PI is an extension of T10-DIF/T10-DIX defined in the NVMe standard. The DIF and DIX configurations as described above are supported, but an additional optional customer-defined metadata block can be added. This block can be placed before or after the DIF/DIX-defined field containing up to 64KB of information. Panther III supports all configurations of T10-DIF, T10-DIX, and NVMe PI.

The advantage of these fields is that they are always in clear text (unencoded) and can therefore be checked at any time in the data flow. For example, the host can verify them, or the storage fabric, the RAID controller, the NVMe drive, among others. If they prove invalid (possible data corruption), the transaction can fail and be retried, ensuring data protection and resilience through the entire data lifecycle.

Panther III supports the following capabilities:

• Check/verify all fields of the source data submitted for encoding.
  • If submitted in DIX format, Panther III can optionally insert the fields into the data to use for subsequent decode operations.

• Generate and insert (DIF format) or generate and return (DIX format) fields for fully encoded data.