GridFS and image size

[Sumita barahmand]

Hi,

I was trying to quantify the performance of MongoDB relative to the image size store in it, so I ran the three following experiments and measured the throughput for the experiments:

1. Inserted 10,000 images of size 2KB, verified the storing of the images with the appropriate size and then had 1200 java threads (instantiated with 4 java processes) to retrieve these images randomly.

2. Inserted 10,000 images of size 12KB, verified the storing of the images with the appropriate size and then had 1200 java threads (instantiated with 4 java processes) to retrieve these images randomly.

3. Inserted 10,000 images of size 500KB, verified the storing of the images with the appropriate size and then had 1200 java threads (instantiated with 4 java processes) to retrieve these images randomly.

I observed that all three experiments resulted in the same observed throughput. I tried modifying the total thread count and the total number of images stored but still the performance for the different image sizes was the same.

I also monitored the network utilization for the node which had the mongodb server running on it and it seemed that the network utilization did not change either and was the same for all the experiments (there are a total of 5 physical nodes, 4 for each java process and one for the mongodb server).

Can anyone provide me with an explanation?

Thanks

Sumita Barahmand

http://www-scf.usc.edu/~barahman/

http://dblab.usc.edu/users/bg/

[Sumita barahmand]

Hi,

I realized that querying the image using findOne() doesn't actually retrieve the image bytes and only returns the image metadata.

fs = new GridFS( db, "photos" );

DBObject tmp = new BasicDBObject( "_id" , r.get("imageid") );;

GridFSDBFile out = fs.findOne(tmp);

So we actually need to read the outputstream into a byte[] to get the actual content for the image.

http://www-scf.usc.edu/~barahman/

http://dblab.usc.edu/users/bg/

Sumita Barahmand