Netbooks built super computers ? How ? detailed ?
Submitted by atang1 on Fri, 09/26/2008 - 03:27.
Super computers are built with a wire length specification. Limiting propagation delay of data transfer is the major part of computer design. Then having nodes of cpus doing simultaneous parallel processing is the key to success.
Latest super computers are a laugh, because of the lengthy ethernet wires hanging all over the cleanroom.
For many years we promoted USB bus for eventual super computer design. At the end of any USB connector, there is a cpu or more to do single function embedded device. This leaves the mainframe computer to handle data transfer of data stored. Why compute when you can read the answers from a table of data?
Now comes netbook designs of cpus, drams, flash cards and USB bus. We can make billions of servers on backplanes in relay racks, connected by short USB bus with amplifiers. Each one server is identified by IPv6 address.
Reminded me of Juno.com and the $50 million Intel investment, to get idling computers to compute simultaneously from Juno.com.
Now, it will become a reality when China using Shanghai city planning to build data center towns using such a technology. When all the data centers are connected via USB in edgeQam satellite transceivers, it will indeed be a huge super computer around the world.
Why netbooks use mainframe design ? USB synaptic unionfs ?
When USB moved from 12 mbps to 640 mbps speed, it is ready someday to do 75 gbps edgeQam data transfer. It became scalable by software and DSP(risc and analog cpu) architecture.
Then, extreme miniature integration in semiconductors took advantage of software driver simplification by using USB peripherals(instructions only).
So, mainframe servers are now designed to replace IBMPcs as servers. Blades are merely cpus, drams and hdds plus ethernet. And the next generation may be only cpus and flash cards plus USB.
Linux unionfs(or aufs) will merge with synaptic(IPv6 data source); apt-get to sync all priority regional data with permanent terabyte hdd stored data.
Data centers will never be the same, again? Will raid6 be replaced by Synaptic checksum restore in unionfs? We hope so.
Footnote:
Synaptic and apt-get are both short programs. Unionfs or aufs should be only in drams or flash cards, Mysql should be in terabyte hdds. Data can be synced via USB bus.