Is HDFS suited to large numbers of small files with random access?

[Manuel Simoni]

Hi,

I am looking for a distributed file system for large numbers of small 
files (a few KB each). Maybe the files would be rows in HDFS.

But my access patterns are different from MapRed-style jobs: reading and 
writing files rather randomly instead of sequential processing.

Is HDFS suited to this kind of workload?

Thanks,
Manuel