From the Founder and CEO, Cloudscale Inc.

Bill McColl

Subscribe to Bill McColl: eMailAlertsEmail Alerts
Get Bill McColl: homepageHomepage mobileMobile rssRSS facebookFacebook twitterTwitter linkedinLinkedIn


Related Topics: Big Data on Ulitzer

BigData: Blog Post

Cloud Computing: The Need For Speed

Who Is Going To Build The Low-Latency Cloud for Enterprise Customers?

The spectacular predictions for cloud revenue growth over the coming decade depend critically on the success of cloud providers in persuading major enterprise customers to deploy many of their mission-critical systems in the cloud rather than in-house. Today's cloud users are overwhelmingly web companies, for whom the low cost of cloud computing, together with simple elastic scaling are the critical factors. But major enterprises are different. In looking at the cloud option, their first major concern is security. This is well documented, has been discussed to death in cloud blogs and conferences, and is certainly a real problem. However, I think we can be very confident that the cloud vendors will be able to quickly overcome the concerns in this area. The reality, although not currently the perception, is that many/most in-house systems are no more secure than cloud systems. The perception of cloud security will change quickly over the next couple of years, as more cloud services are deployed. There are certainly no significant technical impediments here. Many of the security technologies used to protect in-house datacenters can be adapted to secure cloud services.

A second major concern for major enterprises is reliability, but here again we can be confident that the perception on this issue will rapidly change, as cloud vendors continue to improve the reliability of their datacenter and network architectures. Again, no major technical impediments here.

In comparison to security and reliability, the third major concern around enterprise cloud computing has received almost no attention so far. That concern is speed, or rather the lack of it in the cloud. In short, the cloud is slooooooooowww, and as multicore moves to manycore, as the cost of super-fast SSD storage solutions plummet, the difference in speed between cloud computing and in-house systems will widen dramatically unless there is a serious shift in the kinds of cloud services on offer from vendors to high end users.

Consider the most basic of all enterprise services, online transaction processing (OLTP). This is at the heart of all ecommerce and banking. Modern OLTP requires that web transactions and business transactions are processed in a fraction of a second, and that the OLTP system has the scalability and performance to handle even the heaviest loads with this very low latency. But it's even more challenging than that. OLTP systems need to be tightly coupled to complex realtime fraud detection systems that operate with the same very low latency. Can anyone see VISA or American Express moving to the cloud as we know it today?

Amazon Web Services is the main cloud provider today. At Cloudscale we are significant and enthusiastic users of the numerous AWS cloud services. The Amazon team have built an architecture that is very low cost and ideally suited to supporting web companies where the transactions involved are mainly serving up data rather than performing complex analysis. For these simple types of website data delivery, storage services such as S3 offer acceptable performance. But what about modern enterprise apps?

The next generation of enterprise apps will revolve around two major themes: big data and realtime. Even the most traditional of enterprise software vendors, SAP, recognizes this. Hasso Plattner, the company's founder, has been saying for the past couple of years that the future of enterprise software will be based on in-memory database processing that can deliver realtime insight on huge volumes of data.

At Cloudscale we have developed a new realtime data warehouse model that enables both realtime ETL and realtime analytics on big data. The model and in-memory architecture is designed to allow complex realtime analysis, filtering and transformation of big data with latencies of at most a few seconds. This contrasts with traditional data warehouse architectures or parallel tools such as MapReduce/Hadoop, where latencies are usually measured in hours.

In contrast to other companies, and in keeping with our company name!, we have initially deployed the realtime data warehouse in the cloud before deploying it on in-house systems. It is offered today as the Cloudcel service, and currently hosted on AWS.

Later in the year will be launching the Cloudscale R1 architecture for in-house deployment. For this R1 "supercruncher" we have an astonishing array of technology choices available at very aggressive price points: manycore chips, 10GigE or Infiniband networking, massive memory sizes, SSD storage solutions with phenomenal bandwidth (up to 1TB/sec at scale). But more important than all of that is that users will not have to go across the internet to get to it.

This is the technology world we're moving into. Will cloud services be able to keep up, to stay in the race? It's clear that today's cloud services are not optimized for this kind of enterprise computing. Will new cloud offerings emerge? Who will deliver them? Will it be the networking vendors or the telcos? Clearly the central problem in all of this is latency. Who is going to build the low-latency cloud for enterprise customers? If it's you, then we would love to hear about it at Cloudscale. We may just have the killer app for you!

More Stories By Bill McColl

Bill McColl left Oxford University to found Cloudscale. At Oxford he was Professor of Computer Science, Head of the Parallel Computing Research Center, and Chairman of the Computer Science Faculty. Along with Les Valiant of Harvard, he developed the BSP approach to parallel programming. He has led research, product, and business teams, in a number of areas: massively parallel algorithms and architectures, parallel programming languages and tools, datacenter virtualization, realtime stream processing, big data analytics, and cloud computing. He lives in Palo Alto, CA.