Unfortunately, after paying $40 to download this paper, I found it was rather poorly written with a number of factual errors about LMDB. E.g. on page 3 they write "Furthermore, LMDB supports write-ahead logging
(WAL) for durability and crash recovery." which is emphatically wrong. LMDB is crash-proof but expressly does not use a log. Txn logs are an administrative burden that LMDB has eliminated in its design.
On page 5 they go on to write "it becomes apparent that LMDB confronts the map size limit, especially for value sizes
around 128 MB." Again this is false, there is no 128MB map size limit in LMDB.
Too bad the #SAE website doesn't provide any feedback mechanism for the papers it charges so much for. A high price to pay for such low quality technical content.
An amusing part - their benchmark results show that #LMDB trounces #RocksDB and #MongoDB in both read and write performance, yet they still write in their conclusion "RocksDB is optimized for high-performance operations, especially for write-intensive workloads, making it suitable for applications requiring fast data ingestion and low latency."
Add comment