Work with IDE Instances
 | Abbreviations Key |
| CPU | central processing unit |
| FLOPS | floating-point operations per second |
| GB | gigabytes |
| GPU | graphics processing unit |
| HISE | Human Immune System Explorer |
| I/O | input/output |
| IDE | integrated development environment |
| ML | machine learning |
| mo | month |
| PID | process ID |
| RAM | read-only memory |
| RDS | relational database service |
| SDK | software development kit |
| TCO | total cost of ownership |
| vCPU | virtual central processing unit |
| VM | virtual machine |
At a Glance
HISE's built-in Jupyter Notebook environment lets you explore data and derive further insights using R or Python. We offer N1 machines at the exploratory size, C2 and N2 machines at small, medium, and large sizes, and A2 instances for small accelerator-optimized machines. To create an instance, choose a machine type and a Docker image. Upload your results and stop the instance at the end of each workday.
Create an Instance
To create an individual IDE instance in HISE, choose your machine type and one of the available preconfigured Docker images.
Choose a machine type
| Machine type | Specs | Cost |
| Exploratory |
| $24.27/mo |
| Small CPU-optimized |
| $121.93/mo |
| Medium CPU-optimized |
| $487.75/mo |
| Large CPU-optimized |
| $1829.14/mo |
| Small RAM-optimized |
| $153.02/mo |
| Medium RAM-optimized |
| $612.10/mo |
| Large RAM-optimized |
| $2448.39/mo |
| Small accelerator-optimized |
| $2681.64/mo |
Your machine selection determines the performance of your underlying virtual machine. More powerful machines are, of course, more expensive, so use only what you need. Machines are designated by series, generation, and processor type:
- Series. Designated by a letter. Indicates performance and computing power. For example, the C series focuses on advanced performance capabilities, whereas the N series focuses on flexible workloads.
- Generation. Designated by a number. Indicates the version. The higher the number, the newer the version. For example, N2 is newer than N1.
- Type. The preconfigured vCPU-to-memory ratio, such as 16 vCPU to 64 GB RAM
Instances of the N1 type are premier Google Cloud Platform instances. They offer the lowest TCO and the smallest form factor while still allowing for instance suspension.
The C2 machine type offers up to 40% better performance compared with the N1 machine type. However, the tradeoff is a possible 4x reduction in runtime for ML processing compared with N1.
N2 machines offer several advantages:
- A 20% price-performance improvement
- About 25% more memory per CPU
- Twice as many FLOPS per cycle compared with N1
- A 2.82x reduction in runtime for ML processing compared with N1
A2 machines are the first instance type to offer an attached NVIDIA A100 GPU. Even at their smallest form factor, they offer 12 vCPUs and 85 GB RAM.
Choose a Docker image
You can choose from a number of preconfigured Docker images packaged with most or all of scientific libraries you need. We offer stable-release images containing dependable versions of third-party libraries, as well as experimental-release images that support newer versions of third-party libraries. These newly released versions might not have been thoroughly vetted by the issuing provider or the developer community.
We recommend you use stable images unless you absolutely need the experimental versions. If any issues arise from using an experimental release containing an unreliable library, file a bug report with the issuing provider.
To suggest a new image configuration, contact immunology-support@alleninstitute.org.
Stop Your Instance
HISE automatically stops instances that remain idle for a long time. |
After each use of your IDE, upload all critical results to HISE, your private scratch space (see Use HISE SDK Methods), or your project store, and then stop the instance. Nothing is saved in memory after an instance is stopped.
Monitor Disk Space Metrics
You can check CPU, memory utilization, disk I/O, and disk usage metrics for the past 15 minutes and for up to 3 days before the current session. Monitoring IDE metrics can help you troubleshoot issues and pinpoint causes of latency.
Troubleshoot Timeouts
The UI might appear to hang during the 5-10 minutes it takes for your instance to boot. A 524 (Timeout) error occurs if your VM exceeds 80% of its memory capacity, or if the root runs out of disk space altogether. This appears to be a transient issue often remedied simply by retrying the cell, although doing so deletes objects in memory. To kill the process, try the following workaround:
1. To find the running process, run htop. If you have no other cells running, you can usually identify it by looking for CPU or memory utilization.
2. Find the PID for this process. It usually appears to the left of htop.
3. To kill the running process, use the PID you found, and run kill <pid>.
4. If the running process doesn't stop immediately, force it by running kill -9 <pid>. (unless you're writing to a file, because the forcing process can corrupt your data).
To prevent the issue in the future, run saveRDS or a similar function in your notebook to save objects in memory. For more troublesome code, run the R shell in a terminal session.
