Startup Sync Computing has devised a hardware answer to the problem NetApp Spot solves with software: how to optimize the use of large-scale public cloud computing and storage.
To update. CEO Jeff Chou positions Sync against NetApp’s Spot. January 14, 2022. Approach SW. 17 January 2022.
It is operating almost in secret, and what we describe here is not based on company announcements. Instead, it is based on a Article by one of its backers: The Engine, a financial backer based at MIT.
Businesses are finding that running hundreds, if not thousands, of cloud computing instances and storage resources costs significant amounts of cash. Complex cloud computing and storage infrastructure environments are virtually impossible to navigate in real time or manage effectively over time, which means cloud customers spend more, much more than they really need. to get your application work done on AWS, Azure, and Google. , etc.
the genius of Spot.io company bought by NetApp lies in recognizing that the software could help solve the problem. Its Elastigroup product provisions applications at the lowest cost, discounted cloud computing instances, while maintaining service level agreements, and with cost savings of 70 to 90 percent.
Now, two years later, a pair of researchers at MIT’s Lincoln Laboratory argue that the problem is getting so bad that navigating the maze of instance classes across time and clouds requires attacking with hardware and software. They say the problem, classified as combinatorial optimization (CO), is analogous to CO problems in the physical world, such as the classic traveling salesman scenario. It is about finding a route for the sales representative between a set of different destinations to minimize the time and distance traveled.
They have applied their expertise in OC algorithms to hardware design, a parallel processing element, to solve the specific problem of optimizing cloud instances more effectively.
Sync Computing was founded in 2019 by two people: CEO Jeff Chou and CTO Suraj Bramhavar. Chou was a high-speed optical interconnect researcher at UC Berkeley and a postdoctoral researcher running high-performance computing optical simulations at MIT. Bramhavar was a photonics researcher at Intel and later a technical staff member at MIT, developing photonics ICs and new electronic circuits for unconventional computing architectures.
His company received a $1.3 million seed round in November 2019 and more cash from an undisclosed venture round in October 2021. The company website provides a taste of what they are doing, stating: “Future performance will not be defined by individual processors but by careful orchestration over thousands of them. The Sync Optimization Engine is key to this transition, instantly unlocking new levels of performance and savings. … Our technology is poised to accelerate scientific simulations, data analysis, financial modeling, machine learning, and more. These workloads are scaling at an unprecedented rate.”
Sync Computing’s Optimization Processing Unit (OPU) has an unconventional circuit architecture designed to deal with when the number of potential combinations (of instances and instance types for a cloud job) is too high for a server to handle. current search and find the best. They say that as the number of combinations increases, your OPU’s performance outperforms that of general-purpose CPUs and GPUs, requiring less time to find the best combination.
THE OPU uses a design mentioned in a 2019 Nature article of the two founders and others, Analog Coupled Oscillator Based Weighted Ising Machine. This describes an “analog computing system with coupled nonlinear oscillators that is capable of solving complex combinatorial optimization problems using the weighted Ising model. The circuit is composed of a fully connected four-node LC oscillator network with low-cost electronics and is compatible with traditional integrated circuit technologies.”
The Ising model is a mathematical description of ferromagnetism in statistical mechanics and has become a generalized mathematical model for handling phase transitions in statistics.
The paper showed that the OPU, an oscillator-based Ising machine instantiated as a breadboard, could solve random MAX-CUT problems with 98 percent success. MAX-CUT is a CO benchmark problem where the solution is to produce a maximum cut (combination of options) no greater than any other cut.
The paper argues: “Solutions are obtained within five oscillator cycles, and solution time has been shown to scale directly with oscillator frequency. We present a scaling analysis suggesting that large networks of coupled oscillators can be used to solve computationally intensive problems faster and more efficiently than conventional algorithms. The proof-of-concept system presented here provides the foundation for realizing such systems on a larger scale. using existing hardware technologies and it could pave the way to an entirely new computing paradigm.”
To update. We now understand that Sync is focusing on software rather than hardware for its initial product and hardware becomes necessary as the problem grows.
NetApp Sync vs. Spot
Chou sent us his views on how Sync’s technology relates to NetApp’s Spot, saying, “Our solution is much deeper technically than theirs, in fact you can use us in addition to Spot (Duolingo already uses Spot). The profits we made for them were above the Spot instances.
“Essentially we implement a level of optimization that goes from the application to the hardware, which is how we can get even more profit. We are not only based on costs, we can also speed up jobs. We let companies choose whether they want to go faster, cheaper, or both.
“We are also cloud platform agnostic, we work with AWS EMR, Databricks etc. Considering that [NetApp’s] data mechanics it’s just Spark on Kubernetes within the NETapp ecosystem.
“In the longer term, our “Orchestrator” product goes into cluster-level programming to perform global optimization of all resources and applications; something that no one else is doing.”
The Sync Computing OPU could better optimize public cloud resources on a large scale, which means faster and lower cost. Also dynamically, beyond the point where conventional server processors and even GPUs give up. It’s early days for this startup, but its area of focus is the core of NetApp’s CloudOps business unit.
Earlier this month data saver cobalt iron said it had been granted a patent covering technology for optimal use of local and public cloud resources. This technology is based on operational and infrastructure analysis and responds to changing conditions; it is dynamic.
We have two established companies highlighting software approaches to solving the CO problem in the public cloud. If they’ve identified a problem big enough that it’s growing, Sync Computing has a good chance of doing so.