-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathNOTES
47 lines (38 loc) · 1.04 KB
/
NOTES
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
Iterator
- Need to always get the next valid record
Extensible hashing
Effect of data size on latency? Fixed throughput, vary data size.
Flat perf versus data size?
Effect of value size?
Maximum write throughput, cleaner stress test
Effect of a background scan (fixed throughput)
Perf tests
- Maximum throughput as a function of data size
- 99.9th percentile with fixed throughput as a function of data size
- Cleaner stress test (maximum throughput writes)
- Cleaner impact
- Background scans
Variables:
- SAS vs SSD
- Data size
- Value size
- Read/write mixture
Die hard suite
- Apply reads, writes from large, shuffled file, shutdown (cleanly or uncleanly), check contents
Notes
- No locking around I/O
- 1 I/O per operation whp
- Adaptive multi-store cleaner
- Linear scan only for disk scans
- No I/O on index resizing
- Very low heap allocation
- Single-threaded mode
Notes
- Large data performance
Design
-
Tests
- 500M 1000 byte values
- Scan/gc rate: 200MB/sec
- Cleaner stress test
- Can do about 100k IOPs/second on viridant drive mmap in c