Help me with my survy

Question

I am doing this survy for my professor- I am at Bethedsa community college. (Currently, our JVM runs about 400% faster than commercial JVM, so this is definitely in research stage. )
I will appreciate your response- I need to get 20-30 responses- before my professor will release my stipend... I am sure you were a student once- So I will really really appreciate your inputs. You do not have to be complementary or negative to the idea- just be unbiased and honest...
We are building a better version of JVM which will allow profiling in production. This technology, if feasible and successful may be ported to commercial application servers such as WebLogic, WebSphere, JBoss and others, with the cooperation of J2EE vendors.
In this document, we will be calling this system- "DiagnoseNow".
The DiagnoseNow project is in research stage, and we want to adjust the goals of the project to better meet the requirements of the application administrator, designer and IT manager.
Your inputs will be tremendously appreciated.
PROFILING AND TRANSACTION TRACKING IN PRODUCTION
We are building a better version of JVM which will allow profiling in production. This technology, if feasible and successful may be ported to commercial application servers such as WebLogic, WebSphere, JBoss and others, with the cooperation of J2EE vendors.
In this document, we will be calling this system- "DiagnoseNow".
The DiagnoseNow project is in research stage, and we want to adjust the goals of the project to better meet the requirements of the application administrator, designer and IT manager.
Your inputs will be tremendously appreciated.
We are giving away one $50 gift certificate at Amazon.com to one respondent. We expect about 25-50 responses only- So your odds of winning are high.
The features of this proposed JVM are best explained using an example of an ecommerce site.
Ecommerce site scenario
Description of system
Consider a typical ecommerce site with the following features: User Validation, Product Search, Shopping Cart Management and Checkout.
Each user that actually purchases or attempts to purchase a product on the website is consider a "Tracking unit" or a "Session". This new research JVM will allow the time spent in a particular "Sessions" to be tracked at the method level.
System administrator can if he / she so wishes get a report with the following columns:
Session ID, MethodName, Time spent in the method(inclusive of called methods), Time spent in the method(excluding called methods)
Put simply, the goal is to support session level tracking and profiling at the method level in the JVM itself.
This will allow faster debugging and application turnaround, reducing application maintenance cost.
A production ecommerce site could have millions of users and many more transactions per day. We are optimistic about supporting real life production sites.
User Validation
1. login: This validates the user using user name and password.
2. logout: This logs the user off.
3. passwordVerification: Using a database, the password is
verified.
4. retrievalOfLostPassword: If the user has lost his/her
password the password is emailed to the user.
Product search
1. byname: Search product catalog by name.
2. byDescription: Search product catalog by Description
3. bySku: Search product catalog by SKU
4. byBrand: Search product catalog by brand
5. Shopping Cart
6. add: Adds items to the cart
7. remove: Removes Items from the cart.
8. update: Updates Items from the cart.
9. checkout: Checkout items from the cart, by asking for credit
card. .
10. retrievePastData: Retrieve user information from the past.
11. validateCreditCard: Validate credit Card by checking with
the bank.
12. sendConfirmationEmail: Send Email confirming the purchase.
13. sendConfirmationEmailShipping: Send Email confirming that
the product has shipped.
14. sendSurveyEmail: Send Email checking customer satisfaction.
What is a "Transaction" or a "Tracking Unit" or a "Session"?
Each user who searches the product catalog is considered a "Tracking unit" or sessions. A session ends when the users leaves the website.
Transaction tracking/ reporting data at method level
Time spent by each user, in each and every method above is tracked in two ways: Time spent in a method itself, Time spent in the method
and the called methods. This will be done for each and every invocation of the method.
Email alerts
Emails can be generated whenever a particular method(eg.
sendSurveyEmail) takes longer than a specified threshold.
Analysis Reports
Analysis can be done on a variety of topics: Reasons for abandoning shopping carts, Slow or underperforming parts of the application.
Thread dumps
If a particular method runs slower than a threshold value, then optionally a thread dump may be taken and stored. In this way, even if the slowdown occurs at midnight, the thread dump will still be available.
Triggered heap profile
Similarly, if the system is running out of heap memory a heap profiler may become active. Heap profile may also be taken periodically, to allow analysis of heap growth.
The system is best explained using a few examples:
Slowdown in search-books
Problem: Analysis of sessions shows that a large number of customers are browsing for books, but conversion to actual sales is slow.
Analysis: The book database has grown in size, and the server running the database has many more apps on it.
Moving the database to another machine solved the problem.
How and Why did DiagnoseNow help?
DiagnoseNow maintains history of all sessions. The history was analyzed to show customers who were not converting to purchases- a large proportion of them were searching for books, and abandoning the site afterwards.
Slowdown in search-general
Problem: Search has slowed down for all products.
Analysis and Resolution: Analysis showed that all search methods are taking 250% longer than normal. The slowdown happened after a specific date- it was the date on which the OS was upgraded. Rolling back the upgrade and reinstalling it properly resolved the issue.
How and Why did DiagnoseNow help?
DiagnoseNow maintains a baseline of past search method response times. So when a slowdown happens, there is no time wasted "apportioning blame" or debating whether a problem exists.
Credit card authorization- intermittent failure
Problem: Credit card authorization was failing intermittenty after hours
Analysis/Resolution: Credit card authorization was failing intermittently. The automatically generated thread dump showed a faulty connection. The problem was traced to the credit card authorization end, and was resolved.
How and Why did DiagnoseNow help?
Without the thread dump this problem could not have been solved, and without DiagnoseNow the problem would not have been detected unless the system administrator was able to take thread Dumps- for that to happen the problem would have had to occur during normal business
hours. (The off hours system administrators are Unix technicians with no app server or Java knowledge. )
Slowdown of overall system at midnight
Problem: The entire system slows down at midnight.
Analysis/Resolution: All methods show a slowdown around midnight, degrading system performance.
How and Why did DiagnoseNow help?
DiagnoseNow detected the slowdown, and it was discovered that a large admin program ran at midnight. Splitting up the work in the admin program into 5 chunks, made the performance impact lot smaller.
Slowdown at 9.05 pm Saturday/Sunday evening
Problem: The entire system performance degrades by 50% around 9.05 pm on Saturday/Sunday evening
Analysis/Resolution: The weekend cleaning crew was unplugging one of the appservers, to plug in the vaccum cleaner.
How and Why did DiagnoseNow help?
DiagnoseNow allows method level historical tracking and system level analysis. This allowed quick detection of an overall slowdown. Without system level analysis, a variety of alerts may get triggered, but the root cause may not be identified.
SURVEY
Your Name(Required to get the prize):
Your role: Developer/IT Administrator/Manager/ App server
administrator/Software QA
Email address(Required to get the prize):
Phone number(optional):
Place a tick mark against the question:
1. Reducing application maintenance cost is important to me.
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree
2. Triggering a thread dump based on specific conditions is an
important feature
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree
3. Tracking each and every session/transaction down to the
method level is an important feature
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree
4. Rapidly localizing and diagnosing a problem is important to
me
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree
5. Creating alerts based on overall transaction performance is
important to me
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree
7. I can accept a CPU overhead of 8% for extensive monitoring leading to reduced costs.
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree
8. If DiagnoseNow is proven to be a stable system, then I will be willing to pay 1500 dollars per CPU license fee(inclusive of 2
business day email support. )
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree
9. If DiagnoseNow is proven to be a stable system, then I will be willing to pay 7500 dollars per CPU license fee(inclusive of
support with 1 hour response time)
Disagree Somewhat disagree Neutral Somewhat Agree
Strongly Agree

Answer 1

Profiling in production is a pipedream.
With modern JVMs and processors, hundreds of thousands of methods will be invoked on multi-CPU machines in just one second. Tracking and analyzing this data is beyond the power of the fastest database.
To analyze one hour of data, 360 million methods will have to be handled for a 4 CPU Pentium/Xeon machine.
This is clearly impossible to do in real life, with current hardware. May be 5 to 10 years from now, things will be better- however, by then CPUs will be processing many more methods per second...
I think it is a nice ACADEMIC project!