Tuesday, February 22, 2011

Problems with HDF5

In my earlier post, we did some writing to the dataset and that too variable length data by extending the dimensions of dataset and doing this in a loop.

Hmm, let me tell you, it is never easy to work with any open source utility or technology. Because, you never know what can go wrong. So I always reckon Murphy's laws.

At this point of time I was relaxed and very happy that my job has been done. But, when it came like a boomrang to me with higher CPU usage and memory leak issues I could not stopped myself from throwing into tears.

After discussing it with hdf-people I came to know that, hdf5 version 1.8.5 the one which I am using has some serious memory leak issues and CPU usage goes on increasing as if you run it over the time. Some of the members from hdf5 group are working on this issue and is expected to be resolved in hdf5 version 1.8.7 (May 2011).

One more problem was with 'H5DWrite' API which we have used to write into dataset. This APIs is also having the same problem when it is called in a thread/loop.

Huuuh, I was literally strolling with HDF5 and needed some approach which will work for me in its first go. Here, one more data structure sparked in my mind was 'Packet table'.




2 comments:

Santo said...

I will be updating this blog soon with more details about packet tables.

Anonymous said...

Could you please demostrate it soon so that I'll use it.