Skip to content
About
Main content
Model

Scan time with per-file overhead

The first term is overhead: it grows with the number of files, not with the data.

Files
files the scan opens
Per-file cost
fixed per-object cost (round trip, footer, metadata), 20–50 ms on an object store
Parallelism
how many files are fetched at once
Data
bytes of data read
Bandwidth
aggregate read bandwidth

Recurring principle: Granularity mismatch

How it behaves

One input on the horizontal axis, the others held fixed.

Data table
Scan time for 10 GB by file size (F = 50 ms, 32 fetches at once, 1 GB/s)
File sizeScan time
97.7 KB166 s
110 KB149 s
123 KB134 s
138 KB121 s
155 KB109 s
174 KB97.9 s
195 KB88.3 s
219 KB79.8 s
245 KB72.2 s
275 KB65.4 s
309 KB59.4 s
346 KB54 s
389 KB49.2 s
436 KB45 s
489 KB41.2 s
549 KB37.8 s
616 KB34.8 s
691 KB32.1 s
776 KB29.7 s
870 KB27.5 s
977 KB25.6 s
1.07 MB23.9 s
1.2 MB22.4 s
1.35 MB21.1 s
1.51 MB19.9 s
1.7 MB18.8 s
1.9 MB17.8 s
2.14 MB17 s
2.4 MB16.2 s
2.69 MB15.5 s
3.02 MB14.9 s
3.38 MB14.4 s
3.8 MB13.9 s
4.26 MB13.5 s
4.78 MB13.1 s
5.36 MB12.8 s
6.02 MB12.5 s
6.75 MB12.2 s
7.58 MB12 s
8.5 MB11.8 s
9.54 MB11.6 s
10.7 MB11.4 s
12 MB11.2 s
13.5 MB11.1 s
15.1 MB11 s
17 MB10.9 s
19 MB10.8 s
21.4 MB10.7 s
24 MB10.6 s
26.9 MB10.6 s
30.2 MB10.5 s
33.8 MB10.4 s
38 MB10.4 s
42.6 MB10.3 s
47.8 MB10.3 s
53.6 MB10.3 s
60.2 MB10.2 s
67.5 MB10.2 s
75.8 MB10.2 s
85 MB10.2 s
95.4 MB10.2 s
107 MB10.1 s
120 MB10.1 s
135 MB10.1 s
151 MB10.1 s
170 MB10.1 s
190 MB10.1 s
214 MB10.1 s
240 MB10.1 s
269 MB10.1 s
302 MB10 s
338 MB10 s
380 MB10 s
426 MB10 s
478 MB10 s
536 MB10 s
602 MB10 s
675 MB10 s
758 MB10 s
850 MB10 s
954 MB10 s

Where it appears

Each mechanism states it in its own names. The general symbol each name stands for is shown with it.