Intel's Crystal Forest platform promises equivalent or better performance from standard Intel chips compared to network-specific ASICs.
If there was any doubt that Intel is looking to strengthen its presence in high-end networking, that's been removed with the announcement of a next-generation communications platform dubbed Crystal Forest.
Intel has long had a presence in the data centre, and not just in CPUs. The company has been producing network interface cards and switching hardware for years, but finds itself very much a bit player in a market dominated by the likes of Cisco and Bay Networks.
The first clue that Intel was looking to change that came when it announced the $125 million purchase of
InfiniBand technology from QLogic. Designed to provide high-bandwidth, low-latency network connections in a supercomputing or high-performance computing environment, InfiniBand is a technology which could prove key to Intel's efforts in the field of exa-scale computing.
That purchase could have been a one-off designed to provide an infrastructure for supercomputing systems based around the Many Integrated Cores (MIC) architecture. Today's announcement, however, suggests that it's part of a broader attack on core networking infrastructure by the company.
'
The demand for increased network performance will continue to grow as more smart devices connect to the Internet every day,' claims Rose Schooler, general manager of Intel's Communications Infrastructure Division. '
With the popularity of social networking and other high-bandwidth services, such as video and photo uploads/downloads, interactive video, crowdcasting and online gaming, service providers will be challenged to efficiently provision sufficient upstream capacity and manage the spike in network traffic.'
The solution, Schooler claims, is Crystal Forest. Designed as a next-generation platform for scalable high-speed communications, Crystal Forest promises Layer 3 packet forwarding performance of around 160 million packets per second without the use of application-specific integrated circuits (ASICs) or network-centric specialised co-processors. Instead, Crystal Forest works entirely on existing multi-core Intel processors.
Using a software development kit, the Intel Data Plane Development Kit, the company claims that networking performance on Intel architecture platforms can be boosted fivefold over previous generations. Crystal Forest is also set to include QuickAssist technology, which Intel claims can accelerate specialised networking workloads including cryptography, compression and deep-packet inspection without the need for dedicated hardware.
It's a technology Intel claims is necessary: according to the company's figures, based on the Cisco Visual Networking Index report, 30 minutes of video are uploaded to the internet every minute. By 2015, this is expected to grow to five years of video every second. Using off-the-shelf Intel processors, the company claims, makes a platform based around Crystal Forest more scalable than anything that has come before.
Intel plans to release Crystal Forest later this year, with a simulated implementation available for development use via
Wind River.
6 Comments
Discuss in the forums ReplyArs Technica had an excellent article a week or so ago on massive storage (Facebook, Amazon, etc.) Someone from Amazon stated they add as much storage space to their systems (mostly their cloud) every day as the entire operation ran on back in '01. Granted, that 365-fold in 11-ish years, but look at the the curve of technology itself - it's a logarithmic scale. So, maybe this was a mis-statement/typo, but I'd call it possible. Look at the ridiculous number of pics posted to FB every day. "Here's the bacon I'm having for breakfast", "Here's a pointless pic I took from the train on the way to work", "Here's 40 shots of me and my BFF making duckface in my dirty bathroom mirror" (seriously - can't one single person at least bust the Windex out before taking pics in the mirror?) and on and on.
I'd just like to see the source of the numbers.
Put it this way. If those numbers were correct, at any one point in time, 5-6% of all people with internet access worldwide would be uploading video to the internet.
If you make the assumption that worldwide, about a third of the people are asleep at any one point in time, putting the estimate up to about 10% of awake people uploading stuff to the internet.
So, we thus realise that the article concludes that by 2015, we will each spend about 6 minutes per awake hour uploading video to the internet.
I'm pretty sure the article meant 5 hours a second. Much more manageable, and realistic.