-
|
The performance is 5x slower comparing the regular db (in SSD raid 10) when run with 1 Virtual user . The performance is similar (but still slower) if I increase virtual users to 64. Is this expected? |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 2 replies
-
|
Issue, converted to discussion Q&A. A quick test (on an older PC) with results shown as below from HammerDB v4.3 shows that this is not the case - with the in-memory database returning 1.35X higher performance than on-disk for a single Virtual User. Note that the organization of in-memory and a regular database is different, and this is covered in the HammerDB documentation: You should use SSMS and generate the in-memory reports as described in the documentation to understand what impacts on in-memory performance. In particular, note the hash bucket multiplier and related reports depending on the number of tests you run. On-disk |
Beta Was this translation helpful? Give feedback.
-
|
May I know how many warehouses did you configure and what is the hash bucket multiplier ? My test is based on 800 warehouses and the TPM is 20% of your testing. For the on-disk one, did you test in HDD instead of SSD? |
Beta Was this translation helpful? Give feedback.
-
|
Yes, 10 warehouses is sufficient for testing 1 virtual user https://www.hammerdb.com/blog/uncategorized/how-many-warehouses-for-the-hammerdb-tpc-c-test. Also clearly for in-memory you need to allow enough memory for the schema configured. The disk is an SSD with a Haswell (2013) era CPU. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you. May I know why I see big performance degradation for in-memory db when I use 800 warehouses (big db). I do not see such performance degradation for the on-disk db. |
Beta Was this translation helpful? Give feedback.
-
|
The starting point I would recommend is to view the memory allocation and generate the in-memory report as described here https://www.hammerdb.com/docs/ch04s04.html#d0e1563. |
Beta Was this translation helpful? Give feedback.
Issue, converted to discussion Q&A. A quick test (on an older PC) with results shown as below from HammerDB v4.3 shows that this is not the case - with the in-memory database returning 1.35X higher performance than on-disk for a single Virtual User.
Note that the organization of in-memory and a regular database is different, and this is covered in the HammerDB documentation:
See section 3.2 here https://www.hammerdb.com/docs/ch04s03.html
https://www.hammerdb.com/docs/ch04s04.html#d0e1563
You should use SSMS and generate the in-memory reports as described in the documentation to understand what impacts on in-memory performance. In particular, note the hash bucket multiplier and related rep…