Release Notes Change History

NVIDIA Accelerated IO (XLIO) Documentation Rev 3.10.5

Revision

Feature/Category

Description

3.0.2

RX Polling Control

Added support for a new parameter to control RX ring polling inside RX API calls - see XLIO_SKIP_POLL_IN_RX under Configuration Parameter Values.

XLIO Statistics

Added support for Listen socket statistic.

Performance Improvement

Improved performance of SW GRO (generic-receive-offload).

NVME over TCP (NVMEoTCP) Offload

[Alpha] Added support for hardware NVMEoTCP DIGEST calculation offload for TX path at alpha level.

For further details, please refer to NVME over TCP DIGEST Offload Tx (Alpha Level) section.

Bug Fixes

See Bug Fixes section.

2.1.4

Multicast Routing

Added support for reading entries and receiving notifications on local routing tables.

Multipath Routing

Added support for multipath routing through a single hop.

XLIO Statistics

Enabled dump of routing table through xlio_stats utility.

Added listen socket statistic.

Performance Improvement

Improved Tx stability.

Bug Fixes

See Bug Fixes section.

2.0.6

User Space TLS (uTLS) Offload

Added support for uTLS v1.3 offload for Rx path.

SocketXtreme API

Removed support for SocketXtreme API.

Packet Filtering

Removed support for Packet Filtering.

Bug Fixes

See Bug Fixes section.

1.3.5

Connection per Second (CPS) Improvements

Made CPS performance improvements for Hardware TLS offload.

NGINX Crash Recovery

Added the ability to continue forking new child processes in NGINX application even when a crash takes place. The number of new processes that can be created in a crash scenario is set using XLIO_NGINX_WORKERS_NUM.

1.2.9

IPv6 support

[Beta] Added support for the latest Internet Protocol version IPv6.

Note: Support for IPv6 at beta level involves some limitations that can be viewed under Known Issues.

TLS

Added support for TLS v1.2: 256 bits; TLS v1.3: 128 bits; and TLS v1.3: 256 bits.

C++11 Standard Support

Product source code is migrated to C++11 standard requirements.

1.1.8

Security

TLS data-path (Tx and Rx) offload allows the NIC to accelerate encryption, decryption, and authentication of AES-GCM (AES128-GCM, TLS v1.2).
TLS offload handles data as it goes through the device without storing it, but only updating its context. This results in enhanced host CPU utilization and TLS throughput.

NGINX with QUIC configuration support

Added QUIC transport protocol, explicitly designed to support multiplexed connections without depending on a single TCP connection.

Performance Improvement

Added Hardware Large Receive Offload (HW LRO) support:

Large receive offload (LRO) is a technique for increasing the inbound throughput of high-bandwidth network connections by aggregating multiple incoming TCP packets from a single stream into a larger buffer before they are passed higher up the networking stack.

Increased Concurrent Connections maintaining high-wire speed:

Supports vertical scaling, limited by available RAM.

1.0.6

TX Flow: Zero Copy

Added support for zero-copy socket send flag MSG_ZEROCOPY with queueing completion notifications on the socket error queue.

TCP Acceleration (kernel bypass) for NGINX

Added TCP acceleration for NGINX high-performance HTTP server and reverse proxy adaptation.

uTLS – User Space TLS Offload

TLS hardware offload (Transport Layer Security) is a widely-deployed network protocol used for securing TCP connections on the Internet and accelerating TLS encryption.

© Copyright 2023, NVIDIA. Last updated on Sep 8, 2023.