Introduction by Jim Koch -
I’m pleased to introduce our next speaker.  Al Hoagland just spoke on the magnetic data recording on disk.  The other side of the story is a story that will hear from Juan Rodriguez in regards to data storage on tape. Juan, like many of our speakers, has a long history in this Industry, it goes back 35 years.  He began, this is somewhat IBM alumni party I guess, at IBM some 35 years ago.  Since then he has founded three companies, Storage Technology Incorporated in 1969, Exabyte corporation in 1985 and Ecrix in 1996,  where he is presently chairman and chief executive officer.  Juan also in his spare time teaches at the college of engineering and applied science at the University at Colorado in Boulder.  He’s a fellow of the IEEE.  Juan it’s a pleasure to have you here today to talk about storage on tape.

Rodriguez's Biography

JUAN RODRIGUEZ
 
    It seems like every once in a while I sneak into enemy territory. Seems like I’ve been coming to San Jose for magnetic data recording lectures ever since I started with IBM. This was always my Mecca.  I did start in Pougheepsie NY in 1963. 
     Somewhat after IBM started in tapes back in 1953, they released this tape drive, the IBM 726, and if you wonder what they’re doing with that sheet in the photo, they’re providing a backdrop for the photographer.  I don’t know if you know any of them, or if one of them is in the audience here, but in fact this was quite a project. You heard some of the early history of tape, and a group of engineers in Pougheepsie started about 1949 at the Kenyon House to develop what was to become magnetic data tape technology. It resulted in this tape drive in 1953. There was the 726, and it was probably a prototype of what became the ½ inch tape drive family. There was a 100 BPI, 7 tracks and it held all of about 2½ megabytes of data.  I guess it would have required about 20 to 25 of those reels of tape to hold my presentation today.  But this drive was the basis for our data tape drive industry.
     These fellows had to solve a lot of problems, and I really take off my hat to them.  Because in fact they had to develop not only those long vacuum columns and reliable vacuum motors and blowers, but those reels of tape. They did use acetate based plastic tape that was available for audio in those days. That was replaced a few years latter with Mylar. I was able to experienced the problems with acetate tape about 20 years later when we, at Storage Technology, installed 200 BPI machines at the Social Security Administration. The legacy of those tapes lasted for a long time. And in 1972 the IRS wanted to improve their performance, so they went to 200 inch per second drives, which STC provided.

            Video		     Probably the one issue these fellows really never thought of when they came up with this NRZI format. There were 7 parallel tracks, each track was data with a 7th track being the track that guaranteed a transition in every bit group. That transition basically opened a timing window for every bit cell in that character. The window was about ½ a bit cell long, and all the bits had to come in within that window. What happened of course was that this introduced did not deal with the problem of tape skew. 


            Video

     The dynamic skew was a problem that we basically had to live with, and it is exemplified by this animation here.  During the write operation the tape may be skewed, we write the data perpendicular to the tape motion or more correctly, parallel to the head.  As we read back and the tape alignment changes, all of a sudden now all the bits within a bit cell are not read at the same clock time and an error occurs.  This problem was constantly addressed through mechanical means.  Tighter and tighter skew methods were provided for and they all basically resulted in better and better reliability, but as reliability was increased, density was increased, which further compounded the problem. 
     Of course, we all know the pressure of having to come up with more and more, and always working at the edge of the technology.  We increased linear density during the succeeding generations of the modern ½ inch reel.  In 1955, to two hundred BPI. and then again in 1957 to 556 BPI and 800 BPI in about 1962. 

            Video		      Each time these technology changes came about obviously the technology road always looks narrower ahead. We improved the problems of skew, and each time then we then increased the density, which of course resulted in more and more skew problems.  And by 1962, I think the designers of the ½ inch tape drives had seen the end of the road. There was no more "road" after 800 BPI.  And yet there was still pressure to keep increasing densities.  By this time we were able to hold about 18 megabytes on a reel of tape. But the world required more and there was these guys in San Jose who were designing higher density disk drives and these disk drives needed to be backed up.
    So what happened?  I think maybe the second revolutionary step came about 1960.  That was the sync bit. The sync bit said, lets take these NRZI patterns in each track and in every 9th bit lets add a clock bit.  Revolutionary step back.  It was actually first tried no on a ½ tape drive but on a drive that was delivered to the government in early 60’s about ’61, before I got to IBM.  The name of the tape drive was Tractor, and it went on a system called Harvest. It went into a security agency.  This was a 1 inch  tape, it had 10 tracks, and it had this NRZI sync feature on it. In this particular picture we see this byte, which is 10 bits wide. That is data is written for 8 bits and then another bit is written.  This is, I guess, the beginning of run length codes today.  Obviously, we are providing enough clock information to run a phase locked loop that then determines the bit cells in each track. As each bit in each track is detected it’s feed into a skew buffer, here I have a skew buffer that is 3 bits wide, and it’s 4 bits long. 

            Video		     By doing this we basically electronically remove a lot of the issues of skew, replacing what it is a relatively complex mechanical system with a relatively simple electronic system, which is buffering the skew.  So in spite of the fact that the tape is seeking back and forth across those guides, the electronic nature of the buffer track allows the bits within each byte to be aligned. Obviously this was quite revolutionary. 
    The first truly commercial success of this new technology was the IBM 2400. It was released in 1966. It was the first clocked format with 1600 BPI where the method of encoding was phase encoding. This was a very successful product. There was an in-between products between this and Tractor. It was hypertape1, hypertape2, etc. They were extremely reliable machines and extremely expensive.  This nine track product became a very ,very successful product. 
    Basically we addressed the following issues, mechanical tolerances, tape guiding, and data reliability. If we were to look at data interchange before 1600 BPI, it was quite a chancy thing. Between installations, between different computer locations, it was highly questionable whether you could truly interchange data properly. To make sure it could be done, there was a technician, a field engineer at each site, who could adjust or miss adjust  if you will, the machine to read tapes from other locations.  The 1600 BPI design gave us a great increase in reliability, probably close 2 orders of magnitude, especially in interchange.
    So that this barrier of 800BPI basically was blown away. This technique resulted in the breaking through of a significant barrier and led us today to about a 125 thousand bits per linear inch and growing.  This in fact was quite a breakthrough. 
    The major step was going from mechanical complexity to basically simple electrical design, I think I’m showing here a change that even though it electrically looks fairly complex, we all know that electrical systems are much easier to reproduce than mechanical systems, especially as they become more and more complex.  This has been the objective of our design for the 35 years that I have been in industry to basically replace as much as the mechanics, basically the complex mechanics, with the electrical equivalent of their functions. Thereby improving in manufacturing increasing reliability and making things much more available.
    These large footprint drives continue to evolve through 1974.  In 1974 we had some very high performance systems. 
    By this time we had achieved the ability to accelerate tape to two hundred inches per second in 1 millisecond.  That basically translates to over 500 G’s acceleration. 

            Video		     Also, we had tape systems that basically reacted "on demand" to the channel requirements for data. 
    This is not showing very well, but we are in a data tunnel. this represents the channel and here the channel is basically requesting data, the drive reacts instantaneously, providing that data to the channel or when the channel ceases to require the data, the data stops. These were big machines. They required a kilowatt of power, even back then, and of course they measured fairly large sizes 60 x 30x 30, the 30x 30 dimension being what was required to go through a regular door.
    The problem: big sizes and power dissipation and cost. The solution: streaming.
    Here we have more or less a generic reel to reel tape drive.  This is the supply reel here and the tape is threaded through the tape path going around a capstan to a takeup reel. This is the actuator that provides the 1 millisecond start time, very high torque, the tape buffer pockets, minimizing the length of tape that had to be accelerated and the vacuum columns, present since day 1, which basically mechanically buffered the time constants of the reel motors which were in the ball park of 40 milliseconds, to this very quick acting capstan motor which had less than a millisecond time constant.  What did you have to do in order to decrease the size of this tape drive?  Well the reels were a fixed size, so I think the easiest thing to do was basically remove the vacuum columns.  This greatly simplified the tape path.  Now all you have to do is go across the head, and in the process, as you can see here, in shortening the columns we also avoided the use of the capstan, and just got rid of another expensive item.  But in the process we went from this very very quick acting system to a relatively low system performance where now the reels set all the start time requirements.

    A 40 millisecond time constant now was the order of the day. Obviously to improve this we introduced what we call streaming. Streaming is a characteristic of computer system to make sure that data was streaming at a fast enough rate that the drive didn’t have to stop. When a tape drive stops it is kind of clumsy. The advantages of this design obviously greatly reduced power no more vacuum systems and all complexity associated were necessary.  I can still remember one day at Storage Technology when the painters came to paint part of the building at the end of the day we had to disassemble every tape drive because there was paint in every column.  We used broth pressure and vacuum and the paint was condensing every where we didn’t want it to be. 
    This also introduced the idea that possibly by making smaller reels we could reduce the size of the tape cartridges and of course the size of the drives. Even though those cartridges were smaller, by increasing the density we easily make up for the change in capacity which remained constant for a relatively long period of time. From basically from about 1973 when reel capacities reach about 200 megabytes, to 1988 when at Exabyte we came out with 2½ gigabytes.  The standard of capacity for tapes was 200 megabytes. Of course the smaller cartridges now could now be handled much more easily by robots.  The smaller size and all these other things resulted in lower cost. 


            Video

     To explain backhitching, we go back to the tape drive we had before.  Now we have these very slow acting reel motors that to get up to speed take quite a while. (You can see the speedometer there on top.)  The data is transferred at full speed, but when the drive stops, it overshoots the mark and has to back up, and then get going forward to get to the following desired tape location.  This backhitching time obviously introduces a performance hit into the drive. It introduces, some wear and tear in the media so that now the tape path has to improve. It’s mechanics so that it can handle that back forth movement of tape much more easily then before, and therefore increases some of the complexity. 
    Now in spite of all these problems the world basically abandoned those big tape drives from 1974 on and the first tape drive in this streaming process was the IBM 3480.  Now we went from this 10½ inch reel to this 5 inch cartridge, and this cartridge could easily be handled.
    But there was an every increasing pressure to increase performance.  I think we face this in every movement of our lives in the computer world. The need for more and more performance. 
    The disk drive industry at around that time realized that if they didn’t improve the compound annual growth rate from the 30% that had been since the first disk drive, and which tape followed at a 20-30% rate, the semiconductor industry would catch up and replace the magnetic disk as the favorite storage media.  So we saw this increasing rate of growth in magnetic disk storage density at the beginning of the decade, and the growth rate went up to 60%.  Tape had remained in the 20-30% range and as you can see from this chart tape has been losing to disk in this cost effectiveness measure. 
    The main reason for this, there’s been no significant break through in technology.  It just seems that every time there has been an increase in capacity or transfer rate in tape in the last 10 years there has also been an increase in cost.
    We at Ecrix have come up with what we believe is a basically new architecture in recording data on tape, that will be the foundation for tape technology in the future.
    We called it a discrete packet format. It’s based on the same principles as packet switching.  We believe it to be probably as revolutionary as that sync bit was back in 1960. 

            Video		     What it will do for tape is basically to do all those great things that we hope new inventions do.  It will increase capacity.  It will match data rates. It will basically improve reliability significantly, and maybe much more than that it will almost absolutely guarantee interchange at least to the same level of performance that we find when we mention data reliability in the laboratory environment. And last but not least it will significantly improve the cost factors in the drive.

            Video		     This technology is depicted by a single track going by (looks like a worm), each one of those little bubbles there, denotes a packet. Each packet is basically independent of every other packet on that track and on that piece of media each packet has it’s own address.  The data has been initially, during the write process, formatted and broken up into these packets.  Each packet has a full address to it. That address belongs to some space in the buffer and as each packet is read, it goes into the proper space in that packet buffer and once it and all it’s neighbors are assembled they can be read out serially. 

            Video		     In order to guarantee that we read every packet you have to over scan. That is, we have to guarantee that every space, every piece of surface on the media is read.  We tend to basically guarantee that at full speed at least every packet is read once and as we slow the tape down we actually tend to read the packets more and more often, the head maintains its speed as the tape slows down. As we slow down the tape we read each packet more and more times. But we really don’t care how many times we read it, we are only concerned that we read it at least once. 

            Video		     So by doing this with this over scanning operation we also achieve the variable speed architecture which allows us to basically match channel speed.  We can go to full speed at any rate that the channel wants and decrease the channel transfer rate in synchronism with the channel needs. Basically go up and down in speed in synchronism with the channel requirements.
    This basic technology allows us to breakthrough the track density barrier since basically we need no track following during the read operation.  This issue of track following is not a problem anymore. That frees us from a lot of other issues.  The need to mechanically align precisely the different drives in manufacturing. These issue are solved by this system.  The data reliability itself is increased again, our ability to multiple scan and slow down in areas of trouble, we can always slow to get multiple passes on each packet.  But more importantly the issue of interchange, as I mentioned before, just about goes out the window as a design problem. We should be able to get as good reliability in interchange out in the field as we would in the lab. Then of course this channel matching capability would let us perform as well as the system want us. 
    So again we believe that this discrete packet format does for track density what the sync bit did for linear density.
   To basically review, progress has involved 4 major steps: the first tape drive, without which I would not be up here, and maybe that would not be to bad, except I guess I’ve always made my living doing this except for a brief period of time doing disk drives and another period doing optical drives then going back to tape.  The second one was the sync bit, which again allowed the breakthrough in recording densities.  If you’ll recall that first tape drive had a hundred bytes per inch capacity, we’re operating today at 5 million.  So we have progressed from a 100 to 5 million bytes per inch (the thickness of the tape has also decreased an order of magnitude from about 2 mills to about 200 micro inches).  Streaming, basically began the process of making the tape drive smaller.  It’s amazing that, often more than I would have thought, you can still get a ½ inch tape drive and it’s probably the oldest technology in the computer environment today. You can go to Qualstar today for your 9 track 800BPI drive if you want. Who would want that, well I guess enough people to make the business worthwhile. 

            Video		     And last but obviously not least is the discrete packet formats which, again we believe that will bring reliability and much higher track densities, much higher capacities, and at a consistent cost point. That’s where we at Ecrix believe that the future lies.  Any one interested in any more detailed about Ecrix please look it up at http://www.ecrix.com.
Thank you