Hi, I'm Sujata from Intel.
In this video, we talk about what
is new in the latest release of the Data Plane Development Kit,
or DPDK 18.05, and how it can benefit you as a developer.
[MUSIC PLAYING]
DPDK is a set of libraries and drivers for fast packet
processing.
You can convert a general purpose processor
into your own packet forwarder without having
to use expensive custom switches and routers.
Using hardware and software architecture advantages
like multi-core support, user space processing,
and high speed IO, DPDK is able to provide
a significant performance enhancement.
In some cases, there is a 10x increase in performance.
DPDK runs as a Linux freeBSD user level application
accessing the hardware devices directly via Poll Mode Drivers,
or PMD, which includes a number of virtual device drivers.
DPDK supports a large number of CPUs and NIC devices.
The CPUs include Intel, ARM, and PowerPC.
NIC Support includes 1 gig, 10 gig, 40 gig, 100 gig NICS,
and multi-vendor support.
The multi-vendor support includes Intel, Cavium,
Mellanox, NXP, and Virtio.
DPDK also supports crypto devices
in a look-aside design plus compression.
To make it easier for developers,
DPDK comes with a large collection of sample apps
and documentation.
Now let's cover some DPDK 18.05 features.
One of the biggest changes in DPDK 18.05
is in the area of memory management.
The memory in use by DPDK now changes dynamically
as the needs of the application changes.
The initial memory footprint of a DPDK app starting up
will be very small, allowing for faster startup.
But then, as the app acquires more memory
for its data structures and packet buffers,
it will acquire them from the operating system.
Similarly, as the structures are no longer needed and released
by the application, the memory used
is released to the operating system.
This dynamic memory support in DPDK
will help users run multiple DPDK applications
on a single system.
This is because the huge page memory on the system
can be shared between the various processors.
This may make system dimensioning easier
compared to the alternative, which in the past
had each process dimensioned for its worst case memory
footprint.
Another feature added to the 18.05 release
is support for data compression in DPDK.
In its initial releases, DPDK started out only
supporting Network Interface Cards, or NICS.
But over the last few years, its device support
has expanded to cover cryptography via cryptodev,
events scheduling via eventdev, and baseband wireless
via bbdev.
The latest addition to this family
is compressdev, which provides the data structures and APIs
to perform compression on data.
In line with how the API sets are designed,
the compressdev APIs are explicitly
neutral in how they look to support the underlying
functionality.
They allow hardware devices to implement
the API to allow both hardware and software accelerated
compression.
Additionally, Intel optimized code from the Intel
Intelligence Storage Acceleration Library Project
will be available as part of the 18.05 release
with other hardware and software drivers expected
as part of the 18.08 and subsequent releases.
There's much more to learn about DPDK,
so follow the links provided to get additional information.
Don't forget to like and subscribe to the Intel Software
YouTube channel.
Thanks for watching.
[INTEL JINGLE]
Không có nhận xét nào:
Đăng nhận xét