'Compiler Research: The Next 50 Years', by Hall et al, is a manifesto on the future of compiler innovation. The paper starts out as a wishlist of future accomplishments (all of which are too general to ever to be reached) but ends with a reasonable set of recommendations. The main point being that compiler technology is still an integral part of computer science, but unlike the rest of the field, the scope and complicity of compiler technology significantly limits the contribution that an… Continue

The Hall paper was published shortly after the 50th anniversary of IBM's release of the first optimizing compiler and gave insight into the needs and trends of the next 50 years. I liked the statistic given early on which indicated the ~$60B or 0.6% of the U.S. GDP is spent on correcting software defects; that is a good indication of the need for better tools to identify program bugs. One of the main points addressed both early and late… Continue

Continue

Internet-scale applications cannot sustain current trends in growth without significants power efficiency improvements. The power efficient Atom processors offer an alternative to the high-end Xeon processors, but with a significant cost to performance. This leads to an interesting question: to appease the data center customers, should Intel focus on making its Xeon processor more power efficient or improving the performance of the Atom processor? Unfortunately, the paper does not provide a… Continue

I want to point out that being non-SQL does not mean that a database is non-ACID. Note that BigTable uses Chubby as its locking manager, which should ensure consistency. BigTable relies on the Google File System, which stores at least three copies of the data, so data is certainly durable. Atomicity and isolation do not appear to be guaranteed, but so what? If you have 1,000 disks, then atomicity is no longer as big an issue, and isolation seems to fall by the wayside as well. With BigTable, we… Continue

The first paper namely "Parallelizing the Web browser" is mainly motivated by the need for browser improvements especially on hand-held devices and mainly discusses how parallelism may help to achieve the goals of generating Responsive and energy efficient browsers. I personally feel this a very informative paper in the modern day context as many of us are looking for substitutes for laptops and hand-held devices could be the same if we overcome the problem of parallelizing browser access on… Continue

I worked on a BI related application in my last job, and I hate the whole field and don't like to read papers on it. Continue

Application - ASCII - ARPANET ... A - a - A… Continue

Am I going MAD?

“MPI on a Million Processors” by Balaji et al asks an important question: how scalable is MPI? As HPC clusters continue to grow in size to truly colossal shared-memory machines, it is important to consider how the size of the cluster impacts MPI. In the end, the study finds that MPI implementations must continue to pay close attention to the memory requirements, but is ready to scale to a million processor machine . Although, I do not find the experiments to be particularly rigorous… Continue