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: CHANGELOG.md
+21Lines changed: 21 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,5 +1,26 @@
1
1
# Cromwell Change Log
2
2
3
+
## 87 Release Notes
4
+
5
+
### Replacement of `gsutil` with `gcloud storage`
6
+
7
+
In this release, all **localization** functionality on the GCP backend migrates to use the more modern and performant `gcloud storage`. With sufficiently powerful worker VMs, Cromwell can now localize at up to 1200 MB/s [0][1][2].
8
+
9
+
In a future release, **delocalization** will also migrate to `gcloud storage`. As part of that upcoming change, we are considering turning on [parallel composite uploads](https://cromwell.readthedocs.io/en/stable/backends/Google/#parallel-composite-uploads) by default to maximize performance. Delocalized composite objects will no longer have an md5 checksum in their metadata; refer to the matrix below [3]. If you have compatibility concerns for your workflow, please [submit an issue](https://github.com/broadinstitute/cromwell/issues).
[0] Tested with Intel Ice Lake CPU platform, 16 vCPU, 32 GB RAM, 2500 GB SSD
17
+
18
+
[1][Throughput scales with vCPU count](https://cloud.google.com/compute/docs/disks/performance#n2_vms) with a plateau at 16 vCPUs.
19
+
20
+
[2][Throughput scales with disk size and type](https://cloud.google.com/compute/docs/disks/performance#throughput_limits_for_zonal) with at a plateau at 2.5 TB SSD. Worked example: 1200 MB/s ÷ 0.48 MB/s per GB = 2500 GB.
21
+
22
+
[3] Cromwell itself uses crc32c hashes for call caching and is not affected
0 commit comments