Blog: Dan E. Linstedt

« Operational Data Warehousing / Active Data Warehousing | Main | IT Agility and Business Stove-pipes »

Fabric Systems, MPP, and SMP Compute Power

I recently attended the Oracle Open-World show, it was well put together. The vendor floor was extremely busy. I learned a little bit about some up and coming technology called FABRIC SYSTEMS. There are a few vendors out there (it seems) trying to make a go of it. As a result of learning about fabric based systems, I'm going on a quest to learn more. What I learn, I will share here - because I believe that these types of systems will change the landscape as well.
Fabric based systems seem to have MPP in mind, but allow the individual SMP's units to act together. They appear to be scalable, and high powered.

Let's start with defining a FABRIC of systems:

Gt: From where does the term "fabric" come? Is the metaphor based on how -- like the threads in a real fabric -- the individual machines appear and act as one? HAAR:That's an important part of it: that a large number of machines look and act like one ... a unified fabric of processing power so to speak. It's also a metaphor for the business and technical agility -- flexibility -- that the technology unlocks for customers. In those ways, an application fabric is kind of the antithesis to "big iron."

I've been watching, learning, and reasoning about big systems for quite a number of years. I've seen the notions of Grid, Grid computing, and application power - but up until now (when I learned about fabric) these concepts didn't seem to cross over in to the hardware world where MPP and SMP live. You've heard me espouse the virtues and values of high-powered (BIG-IRON) SMP machines, and you've heard me discuss the values of MPP computing systems. That is: when it comes to database power, scalability and flexibility.

Oracle (a couple years back), and other vendors over the years have tried to bring Grid Computing to the center stage, but sometimes they've succeeded (in very specific configurations) and most times, they've not succeeded in constructing true GRID computing power.

Well, last week I learned about a company called: LiquidIQ (http://www.liquidcomputing.com/home/home.php) They define themselves this way:

Liquid Computing is the only vendor offering a converged computing, networking and broadband system to scalable computing users. LiquidIQ™ delivers sustained performance over scale, is highly reliable, and it is simple – simple to implement, to operate, to scale and to reconfigure. As a result, the lifecycle economics of LiquidIQ are extremely attractive.

IWhat's interesting about LiquidIQ is the fact that they are nothing but a massive computing engine with no local storage... You can hot-swap blade computers with multi-core processors, the network components are all high-speed, custom built interconnects, the backbones and motherboards all have high speed data transfer throughput. It truly gives new meaning to the MPP looking and acting like SMP on big-iron, without the SINGLE back-plane requirements.

On a business note, their technology saves huge amounts of energy (efficient design), increases up-time, and can put as many as 4,000 core's to work in a single rack... That's impressive. I didn't have time to ask them how much RAM could be placed in the box, but their base entry point is around $120,000. If you need all this computing power for large scale database work... then this would be an answer for you. More on the storage, benefits, and other items shortly...

There's a few other players in the market, one is Fabric7 - you can see an article on IT Edge Here or you can visit their web-site: http://www.fabric7.com/products_q160.php

Their definition of Fabric Computing is as follows:

Fabric servers feature a highly scalable and hardware partitionable symmetric multiprocessing complex that can be configured into a flexible fabric of large or small servers. The fabric offers a pool of compute and network resources that can be dynamically provisioned into servers to deploy applications and services as business demands. The fabric can extend across multiple servers within a datacenter or across geographically dispersed sites, with a single point of control.

System Fabric Works:

Founded in 2002, System Fabric Works (SFW) develops and integrates server, storage and network software solutions that provide platforms for high-performance distributed applications throughout the enterprise. SFW uses the phrase "Fabric Computing" to describe the leverage that Remote Direct Memory Access (RDMA) architectures provide to computing, network and storage systems to deliver higher performance, lower latency and less energy consumption. SFW currently supports InfiniBand (IB) and Low Latency Ethernet/iWARP RDMA fabrics.

Yea, so what, why should I care?
I think this is important technology. The business values and benefits are to SAAS providers, but also to massive world-wide consolidation efforts, and of course to the Business Intelligence community. I think the benefits include the following:
* Lower cost, lower power consumption with more computing resources
* Incredible up-time
* Massive virtualization of single platform hardware (what I'm saying here is that the RDBMS engine and the ETL engine and the BI Reporting engine can live on the same hardware, and share dynamically allocated resources, but live in separate logical spaces).
* All of the sudden, database engines like SQLServer2005, and MySQL become a serious player in BIG DATA SPACES.. Why? Because scalability is hardware based, interconnectivity and network problems have been solved by the engineers, and now - hundreds of terabytes is no longer a problem on a single machine!!
* Business can save money, buy not buying "proprietary hardware/proprietary operating systems"

If we thought that MPP vendors had it good, they did - for a very long time. Now, the fabric computing space is going to make software (like Microsoft Cubes for instance) incredibly powerful and scalable - pervasive to the end-user... Appliance vendors will need to watch out for this space as well, and possibly partner with these Fabric vendors, the landscape is already changing.

Now, about the storage... Most of these fabric devices don't have on-board storage. That's ok, I spoke to Brocade recently about their efforts to provide FABRIC SWITCHED STORAGE with dedicated backbones and super-high speed access, they've got FABRIC computing built in to their latest storage SAN with dedicated accessibility, load balancing switches, and so on... I'm sure that IBM, EMC, Fujitsu, and Hitachi also have similar offerings, I'd like to hear about them. All I'm saying is that high-speed storage won't be the bottleneck here.

This could mean GRID COMPUTING (on a hardware level) has finally arrived, and been packaged in a form that is usable by the masses. Fabric based computing is a huge space to watch, and as I learn more, I'll be sharing more.

If you have thoughts or experience in this area, I'd love to hear from you.

Thanks,
Dan L
DanL@DanLinstedt.com

Share: