Skip to main content

Plan your hosting options - CIR or Monthly transfer?

This blog provides an insight to plan your hosting options.

Calculating CIR
What is CIR? In simple terms, it is the quantity of bandwidth measured as Committed Information Rate in megabits. CIR allows you to channel as much web traffic as possible within a given bandwidth. CIR calculation will help you to model your deployment option, whether to avail hosted model, or host the application in one's own dedicated server, or avail hosting provider service for monthly transfer or avail hosting provider service in pro-rata basis.

What are the important parameters one needs to consider?
One important parameter is, total number of simultaneous sessions that your web application is expecting. With this value, we can derive certain other parameters that would help us in arriving at CIR. Do not mean that simultaneous sessions imply simultaneous connections. This only means that, if we assume each session represents one user, we have that many users using the application at any given point in time.

Next, determine how much time each user likely to spend on each page. This can be a wise guess or some one from business analyst/user testing team will be able to provide. Let's assume, 60 seconds per page.
Now, determine how much time each page will take to load, again it can be a wise guess. Can we say, 10 seconds to load?. If we put the pieces put together, each user during his session, will have the pattern, 10 second page load, 60 second on the page with no traffic, again 10 second page load and so on ...
This means that each session will be engaged in 15% of total (10*100/70 = approx. 15%) time in data transfer per connection/request. If there are such 100 simultaneous sessions, then the application should accomodate 150 simultaneous connections.

Among your target users, 30% (45 of 150)of simultaneous connections have 2 Mbps line (assuming download speed is 1200 Kbps), and the rest 70% (105 of 150) have 512 Kbps DSL line (assuming download speed is 430 Kbps), then it works out be 54000 + 45150 = 99150 (rounding off 100000 Kbps = 97 Mbps). That means to say, You would need arround 100 Mbit CIR.

Apply normal distribution to determine simultaneous sessions
NOTE: For simultaneous sessions to be 1000, the web application in focus will be a huge one. In most of the normal circumstances, consider regular user base and the average session time each user will spend with the system. For example, if we assume, 5000 as regualr user base and each user spends around 20 minutes a day. Now, apply number of users on Y axis and time on X axis (at equal intervals of average session time) and apply normal distribution centered arount peak time, and derive simultaneous sessions. In our case, let's assume this to be 2% of regular user base, which works out to be 100 simultaneous session, finally results in 10 Mbit CIR.


Calculate your monthly transfer
Let's assume that each session contains 10 requests. If each request results in 50 K of datatransfer, then each session results in 10 * 50 KB = 500 KB or 0.5 MB of data transfer. If there are 5000 regular users, then we can imply 5000 sessions in a day, then it results in 5000 * 0.5 MB = 2500 MB approx. 2.5 GB of data transfer. If we assume the system will have this pattern all the 30 days in a month, then it results in 75 GB of datatransfer.

Now, you can decide whether 10 Mbit CIR is worth opting or 75 GB/month of data transfer. Even, plan your hosting enviroment!
Labels:

Comments

Post a Comment

Thanks for your comments.

Popular posts from this blog

GCP: GAE - Memcache best practices

Memcache is a distributed in-memory data cache in front of or in place of robust persistent storage for some tasks. GAE includes a memory cache service for this purpose. Best practices for using memcache: 1. Handling memcache API failures gracefully; Do not expose errors to the end users 2. Use batching capability of the API when possible 3. Distribute load across your memcache keyspace Use sharding and aggregating for improving performance efficiency. Use TTL (expiration policy) to make sure the memcache does not fill-up indefinitely Use getIdentifiable() and putIfUntouched() for managing the values that may get affected by concurrent updates Use batching (getMulti ("comments", "commented_by") ) to fetch related values together instead of one by one Use graceful error handling
Post a Comment
Read more

Innate and Non-innate learning

I am reading a book called 'What did you ask at school today?' by Kamala V Mukunda. Would like to share some learning. The book is intended for teachers as primary audience, nevertheless, good for any adult to gain deeper understanding on learning process. She talks about brain structure, innate and non-innate learning aspects and talks about synergy needed between the two in the first two chapters. Firstly, innate learning is something that would not need explicit training. For example, kids learning the language. They wont feel strained or stressed during this kind of learning, just because they enjoy the process, where as non-innate learning focuses more on class room learning. It is accepted that learning through playful means will have more impact on kids than the impact through the structured learning. A physcologist, David Geary puts it this way - while learning through playful means has more impact, children should be encouraged to learn the skills through structure...
Post a Comment
Read more

Essential GCP services for a new age application

Identity and resource management IAM  Identity aware proxy Resource Manager Stackdriver Monitoring Stackdriver Monitoring: Infrastructure and application monitoring Stackdriver Logging: Centralized logging Stackdriver Error Reporting: Application error reporting Stackdriver Trace: Application performance insights (latency) Stackdriver Debugger: Live production debugging Development management Cloud Deployment Manager: Templated Infrastructure deployment Cloud Console: Web based management console Cloud shell: Browser based terminal/CLI Development tools Cloud SDK: CLI for GCP Container registry: Private container registry Container builder: Build/Package container artifacts Cloud source repository: Hosted private git repository Database services Cloud SQL: Managed MySQL and PostgreSQL Cloud BigTable: HBase compatible non-relational DB Cloud Datastore: Horizontally scalable non-relational (ACID) Cloud Spanner: Horizontally scalable relation...
Post a Comment
Read more