72 
age technology to overcome this limitation it’s worth while to implement systems in 
disk technology with better properties. There are two main ways to improve these 
properties. 
The extremly raise of the access and transfer quality of the devices results to high per 
formance memories, for example IBM 3380 model AK4. This way is connected with 
growing technological difficulties and exponential increasing costs. 
The other way joins the knowledge, that technological limitations can be by-passed on 
the architecture level, with the fact, that the development on the pc-market produce 
inexpensive disks, which follow the high end disks in rapid succession. 
If it is possible to organize an array of inexpensive disks as one unique device a new 
level of access behaviour can be reached by low costs. A theoretical comparsion of the 
IBM 3380 model AK4 with an array of 100 3.5 inch pc-drives Conners CP3100 is 
shown in figure 1. 
A needed array controller not only will satisfy higher demands for organisation and 
control of I/O requests. 
If logical connected dataitems spreaded over all drives of the array the MTTF is : 
MTTFarray = MTTF s ingle disk / array size . 
That means, already arrays of small size haven’t a sufficient reliability. The reliability 
of the array must be improved by using redundant information. The volume for redun 
dant data can be little 
in practice. The enor- 
mous reliability shown 
in fig.l can be realized 
Unit 
High-perfor- 
Array of 
by 10% of check in- 
mance 
100 inex- 
formation. So the con- 
drive IBM 
pensive 
troller must include 
3380 AK4 
disks 
parts to generate 
needed redundant in- 
capacity 
MByte 
7500 
12 
10000 
100 
formation. 
data rate 
MByte/s 
With such a controller 
I/O-rate 
I/O/s 
20 
3000 
it is possible to accom- 
power 
kW 
7 
1 
plish not only a cheap 
cost 
$ 
100000 
100000 
and powerful storage 
cost per MByte 
$ 
13.33 
10 
device but also a de- 
volume 
cub. feet 
24 
10 
vice with a very high 
reliability usually 
reliability 
h 
30000 
820000 
without backup. Fig. 1: Comparison of SLED with RAID Level 5/1/ 
Figure 2 shows the 
basic concept of such an array controller. The controller computer receives orders 
from the host. It generates partial tasks for the other components of the controller, 
supervices and synchronises their work and apportions them data from the total data 
volume. 
A possibility of data pardoning would be into small components (Bit or Byte divided). 
That results in a high data transfer rate and a short response time for big amounts of 
data (images and sequenzes of images). But if a large number of small transfers are 
ordered seek time and rotational latency dominate in the response time of the array. 
That leads to growing response time in practice.