Reviews: Memcache

Is data retrieval from database, killing your performance? Is your database crashing every few weeks, due to load? Does most of your time go into tweaking and tuning the database for performance? Are you spending too much on Db administrator? Do you find Mysql replication an impossible job?

If the answer is yes for any one of the questions above, then you should give a look at Memcache. Memcache is an object caching system. It caches all your db objects to be used by n number of threads or any part of your application or any process across the data centre.

Assuming you have product, which runs on several servers (S1, S2, S3… all of them on the same data centre), you need to optimise it for data retrieval. A Memcache client will have all your key:pair list hashed on it, it can also have all the information about the servers you are using for balancing the load. Now if a request is redirected to one of the servers, S1 and Memcache is installed, then it will look for the object in memcache client, C1. If an existing key pair exists for the object, the obj is returned from the server (any of S1, S2, S3 …) to the S1. If it does not exist, then it will be created and saved in S1 cache.

More documents on the topic.

http://www.danga.com/memcached

http://blog.twitter.com/2008/05/its-not-rocket-science-but-its-our-work.html

http://www.facebook.com/note.php?note_id=23844338919

Reviews: SVM

In the review Post, i am going to write for different tools, technologies, algorithms which are available in opensource community. There will be lot of things common in between the review post. Most common would be their ease of use in Internet industry.

Today’s topic is SVM.

SVM stands for support vector machines. This is a method used for classifying the data and doing regression later on.

Classification: If we look down at Delhi from a high position, what do we see? We see the clusters of houses in Delhi.

What are the properties of people who had formed the clusters? is it economic, social or geographic properties which had decided for the housing of different set of people in Delhi. To classify a data we need to first figure out the dimension on which the data need to be classified. once we find that we plot the clusters on the axis. Now we have information like Cluster A has people with (A(Range1)X,B(Range1)Y,C(Range1)Z) properties. cluster B has (A(Range2)X,B(Range2)Y,C(Range2)Z) properties and so on. Now by classification we mean to find a curve which separate the clusters. To identify the curve we use set of sample data or the test data.

Regression: Once the classification is done on the real data. we do the regression for any new data coming into the system. we input a data with the properties, and the system will give us the most probable cluster the data belongs to.

use of Application: can be used for predictive analysis, like credit ranking, contextual mapping, forecasting in sales and CRM software, optimization techniques and so on.

Download the software from : SVMLight