For achieving maximal runtime performance in f5c (call-methylation and eventalign), performance governing parameters must be selected by the user as explained in f5c-perf-hints. From f5c-v0.5 onwards, preset parameter profiles are provided for a wide range of computer systems from embedded systems to servers. The user can simply specify the parameter profile to f5c with -x option followed by the name of the profile (See tables below) that closely matches the specification of the user’s computer system. Performance governing parameters currently set by a profile are t, K, B, max-lf, avg-epk and max-epk. Note that the user can also provide any of these parameters explicitly in addition to a profile. In such a circumstance, parameters set by -x will be overridden by the user provided parameters.
In the tables below, the number of cores includes hyper-threads in Intel processors. The GPU specification can be ignored if you are using the CPU only version of f5c.
Generic profiles
| Profile name | Description |
|---|---|
| laptop-low | A low-end laptop with a 4-core CPU, 4 GB RAM and a GPU with 2 GB memory |
| laptop-mid (or simply laptop) | A mid-range laptop with an 8-core CPU, 8 GB RAM and a GPU with 3 GB memory |
| laptop-high | A high-end laptop with a 12-core CPU, 16 GB RAM and a GPU with 4 GB memory |
| desktop-low | A low-end workstation with an 8-core CPU, 32 GB RAM and a GPU with 8 GB memory |
| desktop-mid (or simply desktop) | A mid-range workstation with a 12-core CPU, 32 GB RAM and a GPU with 10 GB memory |
| desktop-high | A high-end workstation with a 16-core CPU, 64 GB RAM and a GPU with 12 GB memory |
| hpc-low | A low-end server with a 32-core CPU, 128 GB RAM and a GPU with 16 GB memory |
| hpc-mid (or simply hpc) | A mid-range server with a 48-core CPU, 256 GB RAM and a GPU with 32 GB memory |
| hpc-high | A high-end server with a 64-core CPU, 384 GB RAM and a GPU with 40 GB memory |
Specific profiles
| Profile name | Description |
|---|---|
| jetson-nano | An NVIDIA Jetson Nano with a 4-core CPU and 4 GB integrated memory |
| jetson-tx2 | An NVIDIA Jetson TX2 with a 6-core CPU and 8 GB integrated memory |
| jetson-xavier | An NVIDIA Jetson Xavier with an 8-core CPU and 16 GB integrated memory |
| nci-gadi | A GPU node in the NCI Gadi supercomputer equipped with Tesla V100 |
Notes
- The above profiles are for a typical nanopore dataset and if your dataset properties (e.g., read lengths, events per k-mer) are quite different, the user may still have to follow steps in f5c-perf-hints for best performance.
- The above list of computer systems is not exhaustive. I will keep adding new profiles whenever I get hold of a new computer system. If you wish a new profile to be added for a particular computer system, please open an issue.
- Some of the above profiles are based on values obtained by running f5c-tools. As this method requires a number of iterations and thus very time consuming, some profiles contain interpolated values.
- The parameter values set by a particular profile can be found here.
Acknowledgement
This parameters profile feature in f5c was developed by David Hyland.