Friday, December 22, 2006
Thoughts on Sun Sparc and Grids
Have you ever heard of the ARM chip? It happens to reside in 80% of all mobile devices and has the best watts/mips/dollar performance beating out Sun, AMD, Intel and Azul Systems.
Zzul Systems currently supports 24 Cores and has plans to more than double. Since Azul is a small company using best practices already known by others, it begs the question of why aren't Sun and others making their chipsets as dense. One perspective may hint to these vendors not having much reserve capacity and/or attempting to slow things down by protecting existing product lines. I wonder if any industry analyst can provide clarity on which strategy each vendor is taking.
Getting chip density right is crucial to the success of grid computing. As you are probably aware that the vast majority of enterprise data centers are full. If folks are going to build enterprise applications that consume say 5000 processors then multi-core approaches are the only things that would work.
Another problem not typically discussed is that grid applications sometimes require low latency networks which may become the next IT constraint that needs to be tored down. Folks such as ATT & Level 3 and others have tons of dark fiber in the ground that can be used to reduce latency , but they will not do unless they can figure out how to make a profit. From my seat, using something that is idle for lower margin feels better than simply letting it waste.
The $1 per CPU hour campaign for usage of Sun Grid seems interesting but it also seems like an opportunity for Sun to get some of that dark fiber lit up so that enterprises who have latency constrained applications can take advantage of their grid.
For enterprises who will move lots of data to external grid providers also need to understand that compression algorithms are not there yet. In my own search, I couldn't find many folks outside of the entertainment industry even attempting to invent new ways to compress data which seems like an opportunity that the venture capital marketplace should consider. Minimally, the notion of taking what the entertainment industry has done with compression algorithms such as DIVX, MPEG, etc and figuring out how to apply to enterprise data could become an interesting opportunity. I wonder if BitTorrent can be used for grids and not just for helping us folks who don't pay for closed source software at home to steal it...
Links to this post: