|Make it a great day|
SOLID STATE STORAGE
Ted J. Albrecht
MAKE IT A GREAT DAYE-Mail: Tjalbrecht@cs.com
From April – Nov. From Dec. – March
7097 Montrose Rd 260 So. Sewell Pl.
Woodbury, MN 55125 Tucson, AZ 85748
(651) 735-4356 (520) 885-4364
Hello, February 22, 1999
Regarding our recent phone conversation about Solid State Storage replacing Magnetic Disk Storage.
While I was working at 3M Company in the Information Systems Research and Integration Department (retired December ’94), one of my projects was to research ways to reduce the time and cost to develop and install business computer applications. I investigated: Methodologies; Prototyping; Case Tools; Object Development Tools; Artificial
Intelligence; Expert Systems; Code Generators; Application Packages; etc.. In all situations, the bottleneck was the complexity of how to organize, store and retrieve data to/from magnetic disk storage in order to meet user response times.
I investigated the various Data Base Management approaches and it seemed clear that some type of parallel store/search/retrieve approach, like NCR-Teredata, would be best. The problem was that we would have to utilize multiple data bases of the same data. One for Processing Transactions against, a second for Data Mining, and others for Departmental Data Analysis. The redundancy was necessary in order to meet acceptable response requirements. But the cost and complexity of the multiple Data Bases and the delay in having the most current data available for decision making was not desirable.
I began looking for more innovative ways to overcome the development complexity and inability to have one Data Base. I came upon a company, Zytel, which was projecting the replacement of magnetic disk storage with Solid State Storage devices. The ability to get at data thousands times faster intrigued me. So I began researching Solid State Storage and found some companies that were working on the concept. The major concerns were loss of data if there was a power failure and the cost of Solid State Storage versus magnetic disk storage. IBM solved the data loss concern by announcing cache memory and Expanded Storage (to hold the indexes and the most used records) with battery back up, thus making volatile memory acceptable. Also non-volatile Flash Memory and Ferrielectric Memory chips are available. The cost issue remains, even though Solid State Storage has gone from $300/Mbyte to $3/Mbyte. But, magnetic disk storage went from $17/Mbyte to pennies/Mbyte maintaining the same 18 times cost differential.
I feel the cost issue can be resolved by eliminating the complex Data Base Management approach needed for magnetic disk storage. The following is an excerpt from a paper “Transitions” I recently wrote which addresses Solid State Storage Servers:
Effective and efficient access to data and information is what enables correlations, yielding knowledge. With knowledge, making intelligent decisions is more likely. Decisions arise as choices present themselves. Having confidence in our ability to get to the appropriate knowledge in order to make timely/optimum choices, is our highest priority … it may even save our life/soul.
In order to meet this challenge, I envision a hierarchy of servers. A hierarchy is necessary to support the various processing activities:
Batch: Many current Information System applications are run on some type of scheduled basis - daily, weekly, monthly, etc.
On-Line: Basically this involves a person initiating/interacting with a computer. ATM’s, PC’s and Internet browsing are examples.
In-Line: This involves computers interacting with other computers. Process control is the current biggest user, like a thermostat triggering air conditioning or heating.
There is a storage hierarchy that the servers access to support the above activities. Most data/information is stored on magnetic tape and optical disks, which is used for batch applications. Magnetic disk arrays are used for on-line and in-line applications. Solid state cache is used on a limited basis, applications requiring critical/fast response, due to its high cost ratio to the other storage medias.
As we automate more activities using in-line processing, the need for fast response will grow. I fully expect that solid state memory arrays will replace magnetic disks arrays. And at that point a great AH-HA will take place. That is … even though solid state memory costs more per megabyte, it will actually be cheaper because it will eliminate the need for complex/costly data base management systems, that are needed to facilitate reasonable response time via magnetic disk arrays.
What I envision is servers accessing solid state memory arrays, that use whatever number of transputers necessary to find the needed data/information, from a logical file, to meet the application request time. Each solid state memory array would have whatever number of transputers needed to meet the application(s) response time requirements. The cost of a transputer ($1.00) is small and would vary slightly ($.10/field) based upon the number of fields an application needed to search on for any one search. Additional transputers can be field installed.
Records, containing data, are stored in a logical file in solid state memory array(s) in no particular sequence. A logical file contains a company’s customers, or employees, or products, or etc. Finding a record is accomplished by using a transputer(s), which can search any field or combinations of fields of a record. The only function the transputer performs is finding the desired record(s) and passing the found record(s) address(s) to the input/output device, which will move the record(s) to the application processor. As technology improves, the transputer will pass the found address(s) directly to the application processor, which will process the record(s) ‘in place’ thus eliminating the movement of record(s).
Each request for data will indicate how quickly it is needed and its’ priority level. The search routine will determine how many transputers will be needed to find the desired record(s) to meet the application response requirement. If just one transputer is needed, then the search routine passes the beginning/ending addresses of the logical file to the transputer and the transputer starts its search. If more transputers are needed to meet the applications required response time, the search routine determines the proper number and passes to each transputer its beginning/ending address (a portion of the logical file) and all the transputers, in parallel, begin searching their portion of the file. In essence each solid state memory array would have a pool of transputers awaiting a search request. When a search request is completed the transputer goes back to the pool.
No complex indexing or exotic disk storage schemes. Instead, just brute force electronic firmware, to provide KISS ‘Direct Data Base Management’. And this could lead to eliminating the need to have multiple files for the same data. No longer would it be necessary to have a transaction file, a data warehouse file and departmental decision support files. Using solid state memory arrays with pools of transputers searching in parallel can meet all applications timing requirements going against any one logical file. Adding records is just a matter of checking the logical file that it does not already exist and if not, putting the new record(s) at the end of the logical file. Solid state memory will also eliminate motors needed to rotate the disks in magnetic disk arrays, reducing power requirements. Flash memory and battery back-up will insure against losing data stored in solid state memory arrays if power fails.
Servers accessing solid state memory arrays will be positioned on the network to support required response times. There will be a server in the home to support the many terminals/PC’s/micro processors (temperature control, security, lighting, etc.). The home server will be networked with the owner's vehicle(s) server and with various community servers (schools, library, newspaper, telephone, cable T/V, Wal-Mart, Blockbuster, mall, etc.) that in turn interface with the World Wide Web. Corporate department servers would tie into their corporate server, that in turn would interface with the community servers, which then would connect with the World Wide Web.
Servers would have directories, which would be able to direct a request to the appropriate server or send out a request for a parallel search among a number of servers when necessary.
This hierarchy of servers using solid state memory arrays could automatically tune itself based upon traffic. The tuning could insure that a server could direct the majority of requests to the right server and keep parallel searches to a minimum.
Servers would recommend when and where servers and Solid State Storage arrays should be expanded, added and/or deleted. Also, when a Net-PC should be upgraded to a robust PC.
No longer would the storage and retrieval of data be the bottleneck in developing and installing new Information Technology applications. This will reduce the time and cost of developing systems and reduce operational costs:
- Having to evaluate Data Base Management
approaches and Data Access tools
- In depth analysis of data usage in order to set up
data access indexes
- Normalizing or denormalizing data
Reduce Data Base Administration
- Involvement/approval time and cost
- Staffing and permit remainder to concentrate
on Enterprise Repository
Reduce user training and special staffing costs since access will be simpler
Reduce electrical power usage
- Having to frequently replace mechanical disk
- Expensive license fees for Data Base Management
Systems and Data Access tools and staff to install upgrades
- Staff to balance disk ‘load’ usage
- Staff and computer time to keep multiple files of
same data in sync
Now the challenge is to find the best memory chip fabricators, transputer fabricators and manufacture to integrate the two into the Solid State Disk Array. Also software to create the Direct Data Base Management approach and assemble a team of marketing and sales representatives to get the KISS Direct Data Base Management solution to the Information Technology industry.
Well I think that states what I am trying to accomplish. This is not something I’m doing to make any big financial gain for myself. Rather it is necessary to easily automate most (eventually all) human endeavor to bring about the “Realization Economy”, which will enable humankind to populate the universe. Which is my life’s hoped for legacy. This is explained in the paper “Transitions” which was mentioned earlier.
I would appreciate any comments and if you are interested in joining the team of people I am working with to make this happen.
Make It A Great Day.
Ted J. Albrecht