some operators use useMap and prefix

[staltz]

Before and after of the perf benchmarks:

  migration (using ssb-db)

    ✔ delete db2 folder to start clean
    ✔ ssb-db has finished indexing
-   ✔ duration: 2816ms
+   ✔ duration: 2870ms
    fallback to close

  migration (alone)

    ✔ delete db2 folder to start clean
-   ✔ duration: 2705ms
+   ✔ duration: 2722ms

  initial indexing

    ✔ null
-   ✔ duration: 2008ms
+   ✔ duration: 2043ms

  setup

  key one initial
-   ✔ duration: 454ms
+   ✔ duration: 464ms

  key two
-   ✔ duration: 19ms
+   ✔ duration: 20ms

  key one again
-   ✔ duration: 14ms
+   ✔ duration: 15ms

  latest root posts
-   ✔ duration: 371ms
+   ✔ duration: 364ms

  latest posts
-   ✔ duration: 29ms
+   ✔ duration: 32ms

  votes one initial
-   ✔ duration: 355ms
+   ✔ duration: 383ms

  votes again
-   ✔ duration: 11ms
+   ✔ duration: 0ms

  hasRoot
-   ✔ duration: 302ms
+   ✔ duration: 294ms

  hasRoot again
-   ✔ duration: 12ms
+   ✔ duration: 0ms

  author one posts
-   ✔ duration: 248ms
+   ✔ duration: 249ms

  author two posts
-   ✔ duration: 61ms
+   ✔ duration: 66ms

  teardown

  total:     18
  passing:   18
  duration:  15.3s

[arj03]

Sweet. I had a similar PR here locally. Wanted to get the concurrent update in, but this is good. We should use prefix for key as well. It's about 2x faster. But there is still quite a bit of key collision (around 1/5) so I'll try and maybe do the % trick and see if that helps. Anyway this needs a benchmark update as well, so lets just get this in first and I'll look at key afterwards.

[github-actions]

Benchmark results

Part	Duration
Migration (using ssb-db)	8271ms
Migration (alone)	3515ms
Initial indexing	2471ms
key one initial	764ms
key two	3ms
key one again	3ms
latest root posts	521ms
latest posts	10ms
votes one initial	533ms
votes again	0ms
hasRoot	396ms
hasRoot again	1ms
author one posts	317ms
author two posts	59ms

[staltz]

Sweet. I had a similar PR here locally. Wanted to get the concurrent update in, but this is good.

Oh, I didn't know you had a branch. Sorry about that.

We should use prefix for key as well. It's about 2x faster.

key() operator currently uses prefix indexes (but not useMap), so what makes it 2x faster?

[github-actions]

Benchmark results

Part	Duration
Migration (using ssb-db)	8984ms
Migration (alone)	3789ms
Initial indexing	2457ms
key one initial	816ms
key two	6ms
key one again	4ms
latest root posts	480ms
latest posts	16ms
votes one initial	491ms
votes again	1ms
hasRoot	320ms
hasRoot again	0ms
author one posts	303ms
author two posts	65ms

[arj03]

Don't worry about the branch, it was mostly for testing this stuff.

Key is 2x faster with useMap

[staltz]

Interesting! I thought key() would be like author(), and that author() doesn't benefit from useMap, see your comment here: ssbc/jitdb#95 (comment)

[arj03]

Yeah, changing this line from 0 to 1 makes key with useMap go from 10ms to < 1ms.

[arj03]

But maybe we should add that as an option? prefixOffset?

[staltz]

But maybe we should add that as an option? prefixOffset?

I was typing this right now. :)
I was going to suggest prefixStart but prefixOffset is better.

[arj03]

Cool, will work on that after lunch.