Sunday, May 10, 2009

Hadoop should target C++/LLVM, not Java (because of watts)

Over the years, there have been many contentious arguments about the performance of C++ versus Java. Oddly, every one I found addressed only one kind of performance (work/time). I can't find any benchmarking of something at least as important in today's massive-scale-computing environments, work/watt. A dirty little secret about JIT technologies like Java, is that they throw a lot more CPU resources at the problem, trying to get up to par with native C++ code. JITs use more memory, and periodically run background optimizer tasks. These overheads are somewhat offset in work/time performance, by extra optimizations which can be performed with more dynamic information. But it results in a hungrier appetite for watts. Another dirty little secret about Java vs C++ benchmarks is that they compare single-workloads. Try running 100 VMs, each with a Java and C++ benchmark in it and Java's hungrier appetite for resources (MHz, cache, RAM) will show. But of course, Java folks don't mention that.

But let's say for the sake of (non-)argument, that Java can achieve a 1:1 work/time performance relative to C++, for a single program. If Java consumes 15% more power doing it, does it matter on a PC? Most people don't dare. Does it matter for small-scale server environments? Maybe not. Does it matter when you deploy Hadoop on a 10,000 node cluster, and the holistic inefficiency (multiple things running concurrently) goes to 30%? Ask the people who sign the checks for the power bill. Unfortunately, inefficiency scales really well.

Btw, Google's MapReduce framework is C++ based. So isn't Hypertable, the clone of Google's Bigtable distributed data storage system. The rationale for choosing C++ for Hypertable is explained here. I realize that Java's appeal is the write-once, run anywhere philosophy as well as all the class libraries that come with it. But there's another way to get at portability. And that's to compile from C/C++/Python/etc to LLVM intermediate representation, which can then be optimized for whatever platform comprises each node in the cluster. A bonus in using LLVM as the representation to distribute to nodes, is that OpenCL can also be compiled to LLVM. This retains a nice GPGPU abstraction across heterogeneous nodes (including those including GPGPU-like processing capabilities), without the Java overhead.

Now I don't have a problem with Java being one of the workloads that can be run on each Hadoop node (even script languages have their time and place). But I believe Hadoop's Java infrastructure will prove to be a competitive disadvantage, and will provoke a mass amount of wasted watts. "Write once, waste everywhere..." In the way that Intel tends to retain a process advantage over other CPU vendors, I believe Google will retain a power advantage over others with their MapReduce (and well, their servers are well-tuned too).

Disclosure: no positions

Friday, May 8, 2009

50% of Cloud Compute Power & Carbon Waste Solved by Software Technique

Ever wonder why virtualized servers are usually run at nominal loads of 30..40%? It's because the speed at which VMs can be live migrated is much slower than the speed at which load spikes can arise. Thus, a large "head room" is built into the target loads at which IT is generally comfortable running their virtualized servers.

Faster networking would seem to save the day, but per-server VM density is increasing exponentially, along with the size of VMs (see latest VMworld talks). Thus there are many more (increasingly bigger) VMs to migrate, when load spikes occur.

Besides the drastic under-utilization of capital equipment, it's a shame that the load "head room" is actually in the most efficient sweet spot of the curve. The first VMs tasked on a server are the most power-costly, and the last ones are nearly free (except we only use that band for spike management).

If loads could comfortably be advanced from 40 to 80%, half (or more) of the compute-oriented wasted power consumption, resultant carbon footprint, and excess capital equipment could be saved. And that's exactly what my R&D showed can be done, with a software-only solution. And as an added value, the same technique accelerates VM migrations across data centers, which is where cloud computing is going.

Essentially, we're blowing an extraordinary amount of money and trashing the Atmosphere from out-moded virtualization software. None of the virtualization solutions from VMware, Citrix, or KVM have this new technique yet. And it's not available via Amazon's EC2. Here's to hoping the future is greener. As well, a high-end networking infrastructure throughout will be essential. Time is money, and time to migrate VMs translates to a lot of money! For the next cloud computing decade, "networking is power". Consider getting long cloud providers and virtualization vendors that support accelerated VM migration, and networking vendors that supply the goods. It nearly goes without mention that Cisco has entered the server market...

Please help spread the word, and feel free to bug your providers.

Disclosure: no positions, related patents pending