You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+18Lines changed: 18 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -102,6 +102,24 @@ Tested with HG00154 sample from the 1000 Genomes Project (196M reads, 72GB SAM f
102
102
- LZ4 compression provides the best query performance among all compression algorithms
103
103
- For a 100Mb region query: RNTuple processes **453,736 reads/sec** vs TTree+ZLIB's **197,815 reads/sec**
104
104
105
+
### RNTuple vs. CRAM (samtools) Region Queries
106
+
107
+
Measured on `HG00096.cram` (GRCh37 reference, samtools `view -c -F 2308`) versus the converted RAM RNTuple produced by `samtoramntuple`. Benchmarks run on an Intel i5-8250U laptop with SSD storage.
- RNTuple has higher constant overhead for tiny windows (dozens of reads); CRAM via samtools remains faster for ad‑hoc lookups.
120
+
- For broad regions that touch millions of records, the columnar layout and sparse index give RNTuple a **1.3×–2.6× throughput advantage** over samtools.
121
+
- Large windows (≥50 Mb) saturate sequential access on both formats; RNTuple’s page-level prefetch keeps bandwidth high while retaining exact read counts matching samtools.
122
+
105
123
## TTree Implementation (Legacy)
106
124
107
125
ROOT scripts to convert a SAM file to a RAM (ROOT Alignment/Map) file using the older TTree format and to work with those files.
0 commit comments