254

S. Das et al.

Every test was repeated 20 times and the results were very consistent and in line with

what we expected. The output is shown in Table 2.

We have set the following thresholds for each tier based on the scores and

have tested the following hash algorithms: Spongent-128, d-Quark, Blake2s, and

Keccak256 by various devices of each tier, as detailed in Table 3. Deciding the hash

algorithm for each tier was largely based on the experimental results we got by

testing each tier of devices with 1000 sample blocks per hashing algorithm, which is

discussed in the Result Analysis section. We used our consensus algorithm using a

target difference of 15 for our test with Python3 as our language of choice. Figures 5,

6, 7, and 8 show the created blockchain in all four tiers of devices.

7

Result Analysis

We have created the experimental setup in a simulated plant for IIoT usage with

a host of smart devices with varying computational power, numerous temperature,

and humidity sensors installed in servo motors for monitoring multi-point data and

storing it securely in blockchain to maintain the safety of the plant. The sensor

data are collected for monitoring and help prevent overheating and ensure the safety

and security of the plant. To analyse the efficiency of the implemented model, we

created a statistical comparison of different algorithms per tier and compared our

implementation results to the expectation of this statistic.

To find the optimum hash algorithm for each tier of devices, we have used 1000

sample blocks per tier for each hashing algorithm to find the expected computational

time per block generated, with the exception of non-lightweight algorithms in the

first two tiers, which were taking unfeasible times and hence ten samples were taken

instead. A graph of benchmark scores of the various devices used in the IIoT simula-

tion is given in Fig. 9. The resulting data of the tier-based hash algorithm comparison

are shown in Table 4 and Fig. 10.

In Tier I, we have used the devices Redmi 2 Prime, Raspberry Pi 3B, Moto E cell

phone for sampling data. For this tier of devices, Spongent-128 was providing us

the best computational time per block at 2.91052278 ms, while d-Quark provided a

relatively higher computational time per block at 3.40191614 ms, as seen in Table 7.

More complex hash algorithms such as Blake2s and Keccak256 were taking signif-

icantly higher computational time per block at over 8 min per block, making them

unfeasible for this particular tier of devices at our required target difference.

In the case of Tier II, the devices used for sampling are iPhone6s and Redmi Note

4. In these devices, we are getting very good computational time per block with both

Spongent-128 and d-Quark at 1.44010181 ms and 1.49501915 ms, respectively.

While Spongent-128 is slightly faster, d-Quark provides 80 bits of security over

Spongent-128’s 64 bits. The computational time per block of Blake2s and Keccak256

is still over 90 s in this tier and hence not viable for these devices.

For Tier III, we have used Microsoft Surface Pro 7 and Lenovo IdeaPad Slim

3. From this tier, the disadvantages of lightweight cryptography, primarily its lack