Large Capacity Archival Data Storage

2.5 m Survey Telescope Technical Note 891228-01

This note is primarily a report on our visit to:

We talked to Lou Misshula, a salesman, Dan Gelbart, a senior engineer, and Doug Smythies, a servo engineer.

CREO plans to produce an optical tape storage system with a capacity of 1000 GBytes per tape. The polyester tape is 880 m long and 35 mm wide. Storage is intended to be permanent with a minimum life of 20 years and 100,000 reads claimed by the company. Media cost is estimated to be $5k to $10k per reel. Average access time is 28 seconds; worst case access time is 60 seconds. The drive fits in one rack and average power consumption is 1000 W single phase. It uses a SCSI interface.

Their target corrected bit error rate is 1 in 10^12 (i.e. approximately eight uncorrected bits per reel of tape). They originally assumed a media bit error rate (BER) of 1 in 10,000, but are currently are seeing twice this. If the media is not improved in time, the error correction will be enhanced to compensate.

An important parameter for the survey telescope is the data write rate. The sustained write rate is 3 MByte/second. The imaging portion of the survey project is expected to produce data at a rate of 3 MByte/second. Hence the drive will just do the job if we store raw data, and with a moderate amount of data compression before storage, we will get a comfortable margin.

Data is written on the tape in strips across the width of the tape. Data is buffered in 80 kByte blocks. Three buffers are contained in the 196 KByte buffer memory. As soon as a buffer is full, the contents is written as 32 simultaneous strips, each containing 20,000 bits. In addition, a wide index mark is written (wide so that it can be read at high speed). The data is read back to verify it as it is written, and the strip advanced 50 microns. At full data transfer rates, the tape is stepped 47 times a second, for an average speed of 2.5 mm/sec. 25% of the data written to the tape is error correction information (the true raw speed of the tape drive is 4 MBytes/sec, not 3, counting the error correction information).

The tape transport looks well-designed. The tape follows a simple path, and only touches a few moving parts. Tape tension is controlled, so the tape will not slip and scratch itself on the reel. One disadvantage of the system is the use of glass reels. Glass is required for stiffness, but the result is a fragile reel.

We were concerned about drive reliability. We asked about power failures, drive failures, etc. The engineers claimed that the drive has been engineered so that media damage will not occur due to power failure, etc. The demonstration was a failure, however. When the power was turned off during a high speed search operation, slack accumulated below the takeup reel, was pulled past the read capstan by the tape from the feed reel and torn in two. At least 1 meter was damaged, i.e., 1000 MBytes equivalent. However, they had warned us the test might fail, because they were not sure if that failsafe circuit had been installed yet (the drive was under construction). So rather than drawing adverse conclusions from the failure, we simply suggest repeating the test on a complete unit.

The company is 6 years old. They make another product, an optical PC board mask generator. They have about 20 employees. The company seems to be well capitalized with expensive NC tools, etc. They claim profitability for each year except last year. They currently have orders for 10 drives which they hope to fill by the end of 1990.

They have yet to deliver their first drive which they had expected to deliver last summer. It will be delivered in a month or two. They are currently manufacturing their first pc boards. The first unit will have speedwire boards. Overall, our impression was favorable. The drives were well engineered. Packaging was good. Obvious risks exist with any new company with limited resources, however. This is a new technology which may not work out during production. Even with a successful product, the company could fail due to mismanagement. No second source for the drive exists although CREO has a relationship with Honeywell which may be interested in licensing the technology. The media is currently manufactured by only one company located in Great Britain although Dow Chemical has expressed interest as well.

Magnetic tape should be considered as a backup technology should optical tape not be available. The Exabyte 8 mm drives currently available write 2 GByte at a maximum data rate of 0.25 MByte/second. Digital Review Nov 20, 1989 page 1 describes a new product to be called EXB-8500 available 3rd quarter 1990 at $3,895 for the SCSI drive which will store 5 GBytes. Unfortunately, it is unlikely that the data rate will increase.

Also announced is the EXB-120 CHS, a 'robotic library' for 116 cartridges available 3rd quarter 1990. It is 19" rack-mountable. Four such devices (i.e. four drives with robot library on each) will hold 0.58 Terabyte. The price is $35,000 each without cartridges or drives.

Honeywell has similar videotape system (with dedicated Sun computer) using VHS tape (Digital Review, Oct 9,1989 P. 29). It will be available this quarter at $525,000. The system will handle 600 tapes (3.1 Terabyte) They claim a transfer rate of 1-4 Megabyte/sec sustained.

At the end of 1992, drives storing 5 GByte at 0.5 MB/second should be available. If data can be compressed by a factor of 2 (we feel that this is extremely conservative), four of the new drives in parallel will keep up with the data stream. The entire imaging survey of 8000 GBytes uncompressed will fit on about 800 tapes assuming a factor of 2 compression. Each set of tapes will require 16 m of shelf length for storage. Media cost per set will be $4k at $5 per tape.

About 20 tapes will be needed per night. For reanalysis of the data, if the data rate is approximately the same rate that original acquisition, approximately 50 tapes can be processed per 24 hour day; 16 days would be required. Similarly, copying data to fresh tape or to another media once every five years would require a similar level of effort. With tape changers, the labor needed to load and unload tapes would be quite reasonable. In our view, this is a very conservative scenario. The actual situation should be much better.

With two plausible data storage technologies available, we feel that it is safe to recommend recording all of the ccd data using an information preserving algorithm.