HBase can help one overcome the 'small file' problem of HDFS. But HBase may not be the best methodology once files get beyond a certain size (that size might be in the 10mb+ area). However that is no longer 'small files' and HDFS would be suitable again.
Another aspect to consider is that access for a single row is mediated by a single regionserver, which means if you have extremely high load on only a few rows, you might not achieve scalability. Caching layers (reverse proxies, cdns) can help solve this. But if you are looking at a high volume, high performance download server that can handle millions of clients asking for the same object, you might need to think a bit more. -ryan On Mon, Aug 10, 2009 at 1:13 PM, Tim Sell<[email protected]> wrote: > Yes you can store binary data in hbase, all data in hbase is treated > as just raw bytes. > It would probably suit well things like thumbnails? I think youtube > uses bigtable for the thumbnail images of their videos. And they store > their actual videos in cdn's (and gfs for the longtail I guess). > > So if you have a lot of large binaries, you'll probably want just the > links in hbase, with the files elsewhere. Maybe mogilefs is more > suitable then hdfs though, because hdfs doesn't like lots of "small" > files? > > ~Tim. > > > 2009/8/10 Rocks <[email protected]>: >> Hi,i want to konw whether hbase can store picture and video as blob field? >> Maybe I should store the picture path in it ,and point to the pic or video >> in file system? >> but how to store the pic and video ,how to define the path pointed to the >> file system in top of hadoop? Can anyone give me some suggestion? >> And i am still uncertain where is my hbase storage file in my HDFS ?In the >> command "bin/hadoop dfs -ls" , i cannot see anything related to hbase file. >> It will be greatly appreciated if you give me some hlep. >> >
