A Case Study About Cloud Based Virtual Labs
Poster presentation on 2nd International Conference on Cloud Computing and Service Sciences (CLOSER 2012).
Case study about cloud based virtual labs and corresponding cost advantage in higher education.
1. What Cost Us Cloud Computing?
A Case Study About Cloud Based Virtual Labs
Nane Kratzke
(C)
(A) (B)
Average Box to Maximum Box Ratio
Maximum and Average Box Usage Accumulated Processing Hours per Week
according to Weinman
2000
1.0
50
Average Box Usage
Avg to Max Box Usage Ratio
Maximum Box Usage in an hour
0.8
40
1500
Used Server Boxes
Processing Hours
0.6
30
1000
0.4
20
500
0.2
10
0.0
0
0
13 14 15 16 17 18 19 20 21 22 23 24 25 13 14 15 16 17 18 19 20 21 22 23 24 25 14 16 18 20 22 24
Calendar Week Calendar Week Calendar Week
Is it more economical for practical courses to pro- cloud service provider accounts from Amazon Web Services. Step 2: Determine your dedicated costs
vide classical dedicated educational labs or to use According to the presented figures we can identify different At the Lubeck University of Applied Sciences the procure-
¨
IaaS/cloud based virtual labs? Try to decide for your- phases being more cloud compatible than others from an ment office could purchase the smallest possible server version
self by analyzing the following table of measured real world economical point of view. Training and development pha- (approximately 4 ECU in the AWS universe) for about 3055$.
cost data! ses show very peaky usage characteristics of resources which So our dedicated costs per atomic timeframe (1h) would be
Group Students Project Costs in $ advantages cloud computing. Other phases with less peaky (regarding a 5 year amortization):
A1 5 WRSC Website 88.39$ usage characteristics disadvantage cloud computing.
A2 6 WRSC Website 265.37$ According to Weinmans proof of the “Inevitability of Cloud 3055$ $
d5year (3055$) = ≈ 0.0697 (4)
5 · 365 · 24h h
A3 4 WRSC Website 88.14$ Computing” we used the following maximum variable cost
Step 3: Determine your maximal cloud costs
A4 6 WRSC Website 162.88$ formula as decision criteria.
Equation 1 told us to calculate our cMAX costs in the following
B1 6 Sailbot Tracking 41.17$
way:
B2 6 Sailbot Tracking 57.58$ dAT F (p)
cMAX := (1)
B3 6 Sailbot Tracking 57.46$ at p(TStart , TEnd , uc)
B4 5 Sailbot Tracking 37.42$ This decision criteria can be applied according the following d5year (3055$) 0.0697 $ $
c26w
MAX = = ≈ 1.99 (5)
at p26w 0.035 h h
B5 5 Sailbot Tracking 48.58$ developed four step decision model.
Step 4: Check appropriate cloud resources
Hard – isn’t it? We have answered this question at the Lu-
¨ Step 1: Determine your atp ratio
The following table shows that AWS provides several compa-
beck University of Applied Sciences for several lectures of In our analyzed timeframe 7612 hours of instance usage were
rable instance types with pricings below our maximal costs.
computer science study programs. generated. So the following average amount of servers would
AWS Instance Type ECU Price/h Comparable
The analyzed use case was a college lecture on web techno- be necessary to provide 7612 processing hours within a 26
Micro <1 0.025$ -
logies for computer science students in summer 2011. week timeframe.
Small (Standard) 1 0.095$ -
24x7
Large (Standard) 4 0.38$ o
Training Development phase P M 7612h
avg26w = ≈ 1.74 (2) XL (Standard) 8 0.76$ +
13 - 15 16 - 23 24 25 26 · 7 · 24h
XL (High Memory) 6.5 0.57$ +
Calendar weeks Our maximum server usage within 1 hour was 49 servers in
Medium (High CPU) 5 0.19$ o
In the corresponding practical courses of this lecture students parallel. So we got the following average to peak ratio for a
formed groups of 5 or 6 persons to develop a website for a 26 week timeframe. Conclusions
scientific conference on robotic sailing (project 1) or deve- So due to our peaky usage characteristics virtual labs are
lop a google map based automatic sailbot tracking service avg26w 1.74 more economical for the analyzed lecture (web technologies)
at p26w = = ≈ 0.035 (3)
(project 2) for the same conference. All groups were assigned max 49 than classical dedicated educational labs.
It turned out that virtual labs provide a more than 25 times cost advantage (1/at p26w ≈ 28.73). In average each student ”cost” us 17.27 USD.
(A) (B) (C) (D)
Costs per Month (aligned to Weeks) Main Cost Drivers Costresponsibilty of Groups Histogram of Costs per Group
500
4
GM 5 (6%) GM 4 (4%)
instancehour (62%)
WRSC 1 (10%) GM 3 (7%)
400
3
GM 2 (7%)
Costs in USD
300
# Groups
GM 1 (5%)
datatransfer (0%)
2
adressing (3%)
200
WRSC 2 (31%)
WRSC 4 (19%)
1
100
datastorage (34%) WRSC 3 (10%)
0
0
CW 13 CW 14–17 CW 18–21 CW 22–25 0 50 100 150 200 250 300
Cost Ranges in USD
Acknowledgements. Thanks to Amazon Web Services for supporting our ongoing research with several research as well as educational grants. Thanks to our students and Michael Breuker for using cloud
computing in practical education. Let me thank Alexander Schlaefer and Uwe Krohn for organizing the 4th World Robotic Sailing Championship 2011 (WRSC 2011) and their confidence in our students.