3 ms·

We wanted to use the same setup for all experiments, so we had to choose for an on-disk DB for SQLite, because TPC-DS SF100 catalog_sales does not fit in 16GB m

We wanted to use the same setup for all experiments, so we had to choose for an on-disk DB for SQLite, because TPC-DS SF100 catalog_sales does not fit in 16GB memory.

eatonphil 5y ago
Gotcha. Thanks for clarifying.
masklinn 5y ago
But SQLite is not even present in the TPC-DS benchmark graphs, while Pandas (which the article notes works solely in memory) is… Furthermore the article explicitely says: > We will use customer at SF100 and SF300, which fits in memory at every scale factor.
- lnkuiper 5y ago
  Customer fits in memory, whereas catalog_sales does not. We chose to remove SQLite from the results because it was so much slower. The plots are much less readable when they are stretched out by something that is slower by an order of magnitude
  - masklinn 5y ago
    > Customer fits in memory, whereas catalog_sales does not. Didn't prevent using pandas which had to rely on dynamic swapping? Or is in-memory sqlite unable to use that much memory? > We chose to remove SQLite from the results because it was so much slower. The plots are much less readable when they are stretched out by something that is slower by an order of magnitude So you're using on-disk sqlite because it fits in memory (unlike pandas which also fits in memory) but you're dropping it anyway because it's too slow when it works on-disk?
    - lnkuiper 5y ago
      You are right, we could probably re-run SQLite purely in memory, but only because macos dynamically allocates additional swap. However, I would not expect much improved performance, because I do not believe that SQLite has a different sorting strategy when running in memory. It would only save some i/o operations, which are very cheap on the macbook anyway. Either way, would be an interesting experiment.
ysleepy 5y ago
Have you looked at the https://www.sqlite.org/pragma.html#pragma_cache_size https://www.sqlite.org/pragma.html#pragma_cache_size I think per default it's only 2MiB, increasing it might help. Probably won't beat the other contenders but sqlite has defaults for being a good citizen.

We wanted to use the same setup for all experiments, so we had to choose for an on-disk DB for SQLite, because TPC-DS SF100 catalog_sales does not fit in 16GB memory.

eatonphil 5y ago

Gotcha. Thanks for clarifying.

masklinn 5y ago

But SQLite is not even present in the TPC-DS benchmark graphs, while Pandas (which the article notes works solely in memory) is… Furthermore the article explicitely says: > We will use customer at SF100 and SF300, which fits in memory at every scale factor.

lnkuiper 5y ago
Customer fits in memory, whereas catalog_sales does not. We chose to remove SQLite from the results because it was so much slower. The plots are much less readable when they are stretched out by something that is slower by an order of magnitude
- masklinn 5y ago
  > Customer fits in memory, whereas catalog_sales does not. Didn't prevent using pandas which had to rely on dynamic swapping? Or is in-memory sqlite unable to use that much memory? > We chose to remove SQLite from the results because it was so much slower. The plots are much less readable when they are stretched out by something that is slower by an order of magnitude So you're using on-disk sqlite because it fits in memory (unlike pandas which also fits in memory) but you're dropping it anyway because it's too slow when it works on-disk?
  - lnkuiper 5y ago
    You are right, we could probably re-run SQLite purely in memory, but only because macos dynamically allocates additional swap. However, I would not expect much improved performance, because I do not believe that SQLite has a different sorting strategy when running in memory. It would only save some i/o operations, which are very cheap on the macbook anyway. Either way, would be an interesting experiment.

ysleepy 5y ago

Have you looked at the https://www.sqlite.org/pragma.html#pragma_cache_size https://www.sqlite.org/pragma.html#pragma_cache_size I think per default it's only 2MiB, increasing it might help. Probably won't beat the other contenders but sqlite has defaults for being a good citizen.