I spent two days in Portland getting briefed by Intel’s Data Centre Group (DCG) as part of an influencer program. This article considers how Intel is fuelling the migration to Software Defined Infrastructure (I feel I can say that without any irony or sarcasm) and particularly how networking is a key market for Intel’s growth in the near future.
- Intel is effectively a “forcing function” to drive adoption of its technology in existing markets but encouraging more features, more software than consume more hardware cycles.
- Intel sees networking a major market segment and positioned for large growth.
- Intel is participating in a wide range of open source project and in OEM/ISV projects to drive greater consumption of the x86 silicon in the networking market.
- Intel is promoting new hardware technology that will drive increased consumption/demand in the network.
Software Defined and x86 CPU
Intel laid our a picture that the DCG sees its growth in three areas – cloud, network and big data.
Cloud & Big Data seems fairly obvious and I learned that there is a more going on that I had previously knew about. Take a look at this chart of workloads below and I will make some statements about what I’m seeing here:
- Four market segments – Telco, Cloud, Enterprise and HPC (Tech Comp)
- Public cloud is already 35% of total workloads – higher than I expected.
- I have some uneasy questions about what the concept of ‘workloads’ is measured – only in terms that I would like to understand the data sources and to confirm its interpretation. Its more common to talk about revenue, units or some other business target so its unexpected.
- High Performance Computing (HPC) runs a lot of workloads. I don’t talk about it here but Intel clearly expects this market to grow dramatically and paints colourful pictures about data lakes, video processing and the hedonistic joy of analytics.
- Cloud, Telco and HPC each have few customers making large purchases. Enterprise is large numbers of customers making smaller purchases.
Networking and the x86
Thats a lot of growth. In fact, workload growth at that level is extraordinary. Incumbent networking vendors have sweet dreams of growth like that in their core markets. So how does this play out ?
Intel pitched this transformation and it is summarised in the following slide – in short, Network Functions Virtualization(NFV):
- Intel is contributing to well known open source projects like OpenStack and OpenDaylight to progress cloud orchestration and SDN (which is an enabler for NFV)
- Intel has established a big partner program (ecosystem) that has recently enlisted network heavyweights Cisco & Ericsson.
- Intel is participating in POC trials – surprising but I’m guessing that Intel wants direct feedback and input into how these POCs turn out. Surprising because Intel is strongly partner focused and hasn’t been part of the direct or indirect customer relationship.
- Intel is offering silicon that is specifically designed for this market. The discussion was that the current generation is first efforts and will continue to iterate in the future products.
How Intel Impacts Networking
Many speakers in the DCG group spoke about the internal Intel IT experience of moving to SDI and how implementation over the last few years has changed the service delivery within Intel.
Ok, so Intel has been eating its own dog food and proud of that fact. And the offered data is pretty standard claim fodder for business acceleration “in the cloud” where change reduces from days to minutes except there are nit
I have mentioned the Network Builders Program before:
Intel® Network Builders is an ecosystem of independent software vendors (ISVs), operating system vendors (OSVs), original equipment manufacturers (OEMs), telecom equipment manufacturers (TEMs), system integrators and carriers, coming together to accelerate the adoption of network functions virtualization (NFV)- and software-defined networking (SDN)-based solutions in Telco networks, public, private enterprise and hybrid clouds.
So vendors of all sizes can work together to influence the direction that Intel will take its software. As I understand it, Data Plane Development Kit (DPDK) has been strongly influenced by this forum. Naturally Intel benefits when ISV/OEMs start using x86 to run their firewall, proxy, IDS etc.
Platform Awareness / Workload Placement
I have said before how Intel is pushing hard to keep new features under development. The concept of workload placement is somewhat new. Today, I don’t see this as an Enterprise topic but for Service Providers this is a key topic for NFV. NFV is about x86 VMs that run network applications and think about what happens in mobile base stations, POPs and CPE closets where many custom hardware appliances could be replaced by an virtualization platform with orchestration such as OpenStack.
I will be looking for more information on this Service Chaining topic which is vital to SP/Carrier needs.
Intel is also promoting new hardware technology which have some impact on networking:
- Rack Scale Architecture
- 3D XPOINT
Rack Scale Architecture (RSA)
The basic idea behind RSA is that a server has statically defined CPU, Memory, Storage and Networking. Intel believes that there enough buyers looking to dynamically allocate hardware to build bare metal servers. Imagine a rack with blades of CPUs, blades of memory and shelves of storage plus a high speed bus with a software API that allows for dynamic allocation. Want four CPUs but only 128G RAM ? How about 1 CPU, 512GB RAM and 10x40GbE ports. You get the idea, mix’n’match the hardware you need to run a workload.
I have talked about Rack Scale Architecture on the podcast and I remain uncertain/unconvinced of its value. When would dynamic allocation of hardware assets make sense ? Surely it would be more practical (cheaper, simpler and more useful) to simply have different sized servers available. The lesson I learned from Chassis Switches is that the backplane costs a lot of money – its complex to design, requires a lot of additional silicon components and has a limited deployment making support costs higher.
I did think of one use case that might work in the carrier space where the network edge is looking towards vEPC/vRAN/vPOP where software appliances will replace existing appliances.
3D Xpoint is a manufacturing breakthrough for producing three dimensional NAND that will be used as a storage tier like SSD/HDD. You want it because its faster, more durable and greater capacity than existing SSD technology. Here is the pitch:
It is my view that SSDs & All Flash Arrays have been adding pressure to the Ethernet networks. Use of NFS & iSCSI for VM images for existing arrays need lower latency. All Flash Arrays work better with low latency and more bandwidth. A storage subsystem using 3D Xpoint will be 1000 times faster than SSD and should drive faster Ethernet networks to support the extra data transfer. Some
NVMe is a new storage interface. The following slide sums it up best:
Some thoughts on NVMe from the material presented to me.
- NVMe replaces SAS/SATA with substantial performance improvement.
- Runs at the speed of the PCIe buses and measured in Gigabits per second of throughput.
- reduces the transaction latency of
Note: Intel Optane is the product name for a 3D Xpoint drive but may or not change in the future
Here is the networking dependency, driving more performance and higher speed networking.The following slide shows that increases in network performance promote storage improvements.
The EtherealMind View
This piece has become very long but I am summarising 12 hours of presentations but here are the last impressions I have a week later:
- Intel is participating in the network space and, to some extent, pushing the network market forward to increase bandwidth and reduce latency.
- Intel is, in effect, entering the market for Telcos, Carriers & Service Providers who will use cloud technology at the network edge to virtualise network applainces/services.
- Intel influences the market within open source projects, or with partnerships with vendors. This may not be visible to customers/end users but Intel has a LOT of people working on fundamental technology advances.
I would like to thank for inviting me to excellent event. They paid for my travel and expenses which enabled me to attend this event in Portland, Oregon. My opinions and perspectives remain my own – my disclosure statement is here.
Any errors or omissions are mine and feel free to let me know in the comments below.