OpenStack: The Building Block for Private Cloud

I am back after an extended break from blogging about data storage topics. As I am no longer working at Quantum, I am free to blog about the recent developments in data storage without any concerns.

Last year, I started using cloud servers on Digital Ocean for PeerCube and worked with Amazon Web Services during Coursera's Introduction to Data Science course. Since then, I became very interested in private clouds and methods to establish and manage them.

With the rise of server virtualization, users and applications ability to spin up and spin down pre-built images as needed, and the success of Amazon Web Services (AWS) public cloud, there are fewer reasons for IT administrators to actively manage pooling and allocation of IT infrastructure resources. While there are several commercial (VMware vCloud) and open source platforms (Apache CloudStack, Eucalyptus, OpenStack) that let you build Infrastructure as a Service (IaaS) private cloud, I am particularly excited with the developments happening in OpenStack community.

I was sold on OpenStack as soon as I was able to install a DevStack environment on an old laptop and spin up images within couple of hours. Since then, I have been exploring OpenStack documentation and source code to understand this platform better with private cloud as a specific application in mind.

OpenStack Components

The modularized architecture of OpenStack includes following service components:

1. Compute (Nova)
2. Object Store (Swift)
3. Block Storage (Cinder)
4. Neutron, formerly Quantum (Network)
5. Image (Glance)
6. Identity (Keystone)
7. Dashboard (Horizon)

In the next blog post, I will elaborate on each services components. Being from data storage industry, I am very interested in learning about the storage service components of OpenStack in-depth.

As I realized that I learn best by writing and sharing, future blog posts will be my journey toward understanding OpenStack and how to use it to establish and manage private clouds.

Book Resources

• OpenStack Operations Guide
• Deploying OpenStack
• OpenStack Operations Guide

Lending Club Borrower's Income Verification, Loan Issued Year, and Initial List Status

Recently Lending Club modified the historical loan data file, included several new loan and borrower attributes and removed a few. One of the new fields is whether a borrower's income was verified by Lending Club during the loan application process. Recently, there was a discussion about verified income at LendAcademy forum. Some of the questions and concerns raised during the discussion were:

• Has the number of available loans with verified income gone down recently?
• Does the income verification really matter with loan performance?
• Does Lending Club verify income for all loans?
• Which borrowers are more likely to have their income verified?

It will be interesting to find answers to some of these questions and supporting data from the new historical loan data file.

Loan Issued Year

The chart below shows the borrowers' income verification for the loans by issued year. The loans with verified income are listed as TRUE and with unverified income are listed as FALSE. Two questions listed above are right away addressed from this chart.

• Lending Club doesn't verify borrower's income for all loans issued.
• The percentage of loans issued with verified income has gone up recently. Whether the retail lenders are seeing the loans with verified income at the time of offering is an open question.

In 2007, borrowers' income was not verified at all. Since then, the percentage of loans with verified income has been rising. In 2013 year to date, more loans were issued that borrower's income was verified than the loans with unverified income.

Initial List Status of Loan

Lending Club reserves a few loans for 12 hours and offers them to the institutional and large retail lenders who want to lend the whole amount for a loan. I am not sure whether the historical loan data file includes the loans that were offered and picked up by lenders as 'whole' loans. But, the loans that were initially offered as whole, designated with 'w', but not picked up as 'whole' loans are listed in historical loan data file.

The chart below shows the percentage of loans with verified and unverified income of borrowers with initial listing status and issued year of the loans. Lending Club started offering the 'whole' loans only since late 2012. With the limited data, there doesn't appear to be any significant difference in percentage of loans with verified income between the loans that were initially offered as whole or fractional.

Key Takeaways

• Lending Club is verifying borrower's income for greater percentage of loans issued on its platform recently.
• There is no significant difference in income verification for loans initially listed as fractional or whole.

Lending Club Loans Issued Since 2010 - Principal Paid Back and Months of Payment

This post is the last in the series of posts discussing when default of loans start to peak (Part 1, part 2, part 3, and part 4).

Months of Payment

The chart below shows the percentage of 36 month and 60 month loans defaulted as a function of months of payment for loans issued since 2010. By reviewing both 36 month and 60 month loans issued in same time frame, we may be able to better compare such loans. While the default patterns are very similar for first 10 months of payment, the rate of defaults increases rapidly for 60 month loans after 10 months of payment. 50% of defaults for both 36 month and 60 month loans occurred within 8 months or so, the 80% of defaults for 60 month loans occurred within 13 months compared to 15 months for 36 month loans.

Principal Paid Back

Similar chart for Principal paid back is shown below. It is clear from the chart that while 50% of loans of both maturities defaulted within 8 months, the 60 month loans paid back (9%) only half of principal compared to the principal paid back by 36 month loans (18%). Can the 60 month loans that continue to make payment make up for this extra loss in principal with longer repayment duration and/or higher interest rate?

Another interesting observation from above chart is the increasing difference in principal paid back between loans of 36 month and 60 month maturities. For example, the 20% of defaulted 60 month loans paid back 4% of principal little more than half of 7% principal paid back by 20% of defaulted 36 month loans. In comparison, the 80% of defaulted 60 month loans paid back 16% of principal less than half of 37% principal paid back by 80% of defaulted 36 month loans.

The chart below shows the scatter plot of Principal paid back and Months of payment for 36 month and 60 month loans issued since 2010. A second order polynomial trend line is shown on the chart separately for 36 month and 60 month loans. As the principal portion in monthly repayments for 60 month loans is much smaller than that for similar 36 month loans, the increasing difference between principal paid back with months of payment is understandable.

Key Takeaways

• In the end, the months of payment is much more straightforward method to determine when defaults peak.
• For 36 month loans, 50% of defaults are expected to occur within 10 months of payment, and 80% of defaults within 20 months. 
• The default trend for both 36 month and 60 month loans is very similar for first 8 months of payment.