73 
A better method 
dividing data is 
into components 
match some physi 
cal sectors of the 
DASD. Then a 
parallel work of 
some single disk 
controller on dif 
ferent transfer 
tasks is possible. 
Every single disk 
controller services 
only one disk, 
stores intermedi 
ately the data, de 
tects and corrects soft errors of the data stream. Instead of it an intermediate memory 
pertains to the single controller. 
Fig. 2: Disk Array Controller Configuration 
The correction hardware generates addi 
tional information to the input data 
stream, needed for prospective recon 
structions and restorates destroyed data. 
It is possible to store the needed redun 
dant information on special disks, called 
check disks. Due to several reasons these 
check disks are a bottleneck and deterior 
ate the system performance. Therefore in 
our project we distribute the check data 
across all physical disks. Then the disk 
array is a level 5 RAID system from the 
classification of /1/. 
The connection component links all con- 3- standard IDBMS Configuration 
troller parts. It can be implemented by a 
bus. But there is the disadvantage that 
such a structure couldn’t fit to the needs of the actual transfer tasks. 
In the first step of implementation we will build up a controller for 16 disks connected 
by a modified bus structure. It should work with a reliability of 60,000 h. Performance 
estimations show a transfer rate for large amounts of data from 10 MByte / sec and a 
I/O rate of 300 I/O per second. 
3. Image Database Management System 
The term "Image Data Base" (IDB) is often used with several meanings. The most 
simple IDB is only a collection of images (image archive) - the question how to access 
to these images is not solved. In this sense the above described storage complex with 
high capacity is a well suited device for installing such a simple IDB. It should be 
mentioned that sometimes it is possible to use "intelligent" data distribution methods 
to reduce sub- image retrieval time /2/.