LIBLINEAR is a linear classifier for data with millions of instances and
features. It supports both logistic regression and L2-loss linear SVM using a
trust region Newton method in

C.-J. Lin, R. C. Weng, and S. S. Keerthi. Trust region Newton method
for large-scale regularized logistic regression. Technical report, 2007.
A short version appears in ICML 2007.

Main features of LIBLINEAR include 

Same data format as LIBSVM and similar usage
One-vs-the rest multi-class classification
Cross validation for model selection
Probability estimates (logistic regression only)
Weights for unbalanced data

WWW: http://www.csie.ntu.edu.tw/~cjlin/liblinear/