Intel and the Cloud. Wow.
June 16, 2009
Please, read Dave Asprey’s “What Intel Can Teach Google about the Cloud.” I was surprised. Mr. Asprey wrote:
But these cloud compute providers, liberated from the shackles of Moore’s law, can’t grow network speeds as quickly as they can add servers, creating exactly the same problem that CPU vendors faced when their CPUs grew faster than the system bus. It’s getting worse, too — according to the lesser-known Nielsen’s Law, Internet bandwidth grows at an annual rate of 50 percent, compared with compute capacity, which grows at 60 percent, meaning that over a 10-year time period, computer power grows 100X, but bandwidth grows at 57X. Ouch. So what did Intel and AMD do when faced with the same problem? They looked for a fix they could apply quickly. The quick fix was to add a cache to the processor, which allowed the CPU to run at full speed and store results in temporary memory until they could move across the slower system bus. It also allowed them to keep selling faster processors while they tackled the longer-term project of improving standards for bus speeds.
Three comments:
- Intel tried a cloud play with some expensive data centers and a deal with Convera. Intel demonstrated that it could not deliver.
- Physics is indeed the problem for servers. The Google is dabbling in a range of interesting engineering to compensate for those issues. Caching is one solution, not the only solution. On CPU caches can introduce latency and on die data to’ing and fro’ing reduces some fancy multicore gizmos to piggies.
- Look at the approach of Perfect Search in Orem, Utah. Mr. Asprey’s assessment does not apply to that firm’s engineering approach. Big oversight in my opinion.
Stephen Arnold, June 15, 2000
Comments
One Response to “Intel and the Cloud. Wow.”
-
- Subscribe to Beyond Search
Feature archive
News archive
-
Um…how does Perfect Search’s more efficient search technology equate to a reduction in the need for large pipes to connect to their “dramatically faster” search appliances? They seem to reduce the compute requirement, but my comments about the need to have enough bandwidth to get there still stand. If you read the article, it says that caching solved a problem for CPUs, but that NETWORK caching could solve a similar problem for data centers.
The NYtimes/GigaOM piece said nothing about Intel’s data center project in the 90’s. I met with Intel back then to look at buying their data centers when I was with Exodus, but that was irrelevant to the analogy in the article.