Jürgen Baier wrote:
Hi,
I have a question about LMDB (I hope this is the right mailing list for such a question).
I'm running a benchmark (which is similar to my intended use case) which does not behave as I hoped. I store 1 billion key/value pairs in a single LMDB database. _In a single transaction._ The keys are MD5 hash codes from random data (16 bytes) and the value is the string "test".
The documentation about mdb_page_spill says (as far as I understand) that this function is called to prevent MDB_TXN_FULL situations. Does this mean that my transaction is simply too large to be handled efficiently by LMDB?
Yes.
Hi,
thanks for the answer. However, I still have a follow-up question on this benchmark.
When I add 1 billion key/value pairs (16 byte MD5) to the LMDB database (in a single transaction (but I also get similar results when I add the same data in multiple transactions)) I get the following results:
Windows, without MDB_WRITEMAP: 46h Windows, with MDB_WRITEMAP: 6h (!) Linux (ext4), without MDB_WRITEMAP: 75h Linux (ext4), with MDB_WRITEMAP: 73h
MDB_WRITEMAP seems to have a huge impact on write performance on Windows, but on Linux I do not see similar improvements.
So I have two questions:
1) Could the the difference between Linux and Windows performance regarding the MDB_WRITEMAP option be related to the fact that LMDB currently uses sparse files on Linux, but not on Windows?
2) Is there a way to speed up Linux? Is there a way to pre-allocate the data.mdb on startup?
Thanks,
Jürgen
On 21.11.17 21:17, Howard Chu wrote:
Jürgen Baier wrote:
Hi,
I have a question about LMDB (I hope this is the right mailing list for such a question).
I'm running a benchmark (which is similar to my intended use case) which does not behave as I hoped. I store 1 billion key/value pairs in a single LMDB database. _In a single transaction._ The keys are MD5 hash codes from random data (16 bytes) and the value is the string "test".
The documentation about mdb_page_spill says (as far as I understand) that this function is called to prevent MDB_TXN_FULL situations. Does this mean that my transaction is simply too large to be handled efficiently by LMDB?
Yes.
Jürgen Baier baier@semedy.com schrieb am 29.11.2017 um 14:43 in
Nachricht a157fa5e-c175-7b96-7b9b-046633ce07ea@semedy.com:
Hi,
thanks for the answer. However, I still have a follow-up question on this benchmark.
When I add 1 billion key/value pairs (16 byte MD5) to the LMDB database (in a single transaction (but I also get similar results when I add the same data in multiple transactions)) I get the following results:
Windows, without MDB_WRITEMAP: 46h Windows, with MDB_WRITEMAP: 6h (!) Linux (ext4), without MDB_WRITEMAP: 75h Linux (ext4), with MDB_WRITEMAP: 73h
MDB_WRITEMAP seems to have a huge impact on write performance on Windows, but on Linux I do not see similar improvements.
So I have two questions:
- Could the the difference between Linux and Windows performance
regarding the MDB_WRITEMAP option be related to the fact that LMDB currently uses sparse files on Linux, but not on Windows?
- Is there a way to speed up Linux? Is there a way to pre-allocate the
data.mdb on startup?
It might be worth the efforts to do a block trace to see the access pattern on the block device. There are numerous tuning parameters (block device and file system), but there hardly is a "one size fits all" setting.
Thanks,
Jürgen
On 21.11.17 21:17, Howard Chu wrote:
Jürgen Baier wrote:
Hi,
I have a question about LMDB (I hope this is the right mailing list for such a question).
I'm running a benchmark (which is similar to my intended use case) which does not behave as I hoped. I store 1 billion key/value pairs in a single LMDB database. _In a single transaction._ The keys are MD5 hash codes from random data (16 bytes) and the value is the string "test".
The documentation about mdb_page_spill says (as far as I understand) that this function is called to prevent MDB_TXN_FULL situations. Does this mean that my transaction is simply too large to be handled efficiently by LMDB?
Yes.
-- Juergen Baier
openldap-technical@openldap.org
-
Howard Chu -
Jürgen Baier -
Ulrich Windl