Speeding up simulations

One of the slowest parts of running OghmaNano simulations is not always the CPU computations, but the time spent writing data to disk. Modern processors and memory have become dramatically faster over the last 50 years, but storage performance has lagged behind, especially when compared to RAM speeds. As a result, disk I/O can easily become the bottleneck in resource-hungry simulations.

Why disk access dominates simulation time

Historically, magnetic hard drives were limited by mechanical motion: a head had to be physically moved to the correct position on a spinning disk before any data could be read or written. That seek time is fundamentally constrained by how fast you can move a piece of metal.

Solid-state drives (SSDs and M.2/NVMe drives) remove the moving parts and provide much faster access times, but they are still orders of magnitude slower than accessing data in RAM. Whenever OghmaNano writes large numbers of files – for example during parameter scans, fits, or 2D/3D simulations – this disk access adds up.

As a general rule:

• CPU and GPU arithmetic is very fast.
• RAM access is fast.
• Disk access is comparatively slow, and can dominate runtime if misconfigured.

Choosing the right storage

Different types of storage have very different performance characteristics, and this can significantly affect OghmaNano’s overall speed.

• Local SSD or M.2/NVMe drive
  This is the preferred location for OghmaNano simulations. Writing to a local SSD is fast and has low latency. It is still slower than RAM, but it is usually “good enough” that disk I/O is no longer the main bottleneck.
• Local spinning hard drive (HDD)
  Slower than an SSD, but still often acceptable for moderate workloads. For heavy parameter scans or large 2D/3D simulations, an SSD is strongly recommended.
• Network drives (e.g. home directories on a university server)
  Data must travel across the network each time a file is read or written. Effective bandwidth and latency are typically much worse than a local disk. This can easily slow simulations by an order of magnitude or more, particularly when many small files are created.

In short, for performance-critical work you should always try to keep your simulation folders on a local SSD or NVMe drive.

Cloud storage and automatically synced folders

Many systems now automatically sync user files to cloud services such as OneDrive, Dropbox, Google Drive or institutional backup systems. This is convenient for documents, but can be a serious problem for simulation workloads.

When a simulation is stored in a synced folder:

• Each file written by OghmaNano may be queued for upload.
• The sync client may rescan the directory repeatedly as files change.
• Network latency and server distance add extra delays.

The result is that simulations appear to “crawl”, even on fast hardware, because every file operation is competing with the sync process.

This can be confusing because many operating systems present cloud-synced folders as if they were local home directories. For OghmaNano, it is usually best to:

• Choose a local, non-synced directory for active simulations (for example, C:\oghma_sims\).
• After the simulation has finished, copy selected results to your home directory or cloud storage for backup.

Antivirus and background tools

Modern built-in tools such as Windows Defender are generally well-behaved and do not interfere too much with simulation workloads. However, some third-party antivirus products or security suites are much more aggressive: they may scan every single file that is written to disk, or continuously monitor directories as they change.

Because OghmaNano produces a large number of small files, such scanning can add a significant overhead. Symptoms include:

• Simulations that run much slower than expected compared to YouTube demos or benchmark numbers.
• Disk activity indicators staying constantly busy even when the CPU is not fully loaded.

If you notice this behaviour, you may wish to:

• Exclude your local simulation directory from real-time antivirus scanning (if your IT policy allows it).
• Avoid installing multiple overlapping security tools.

Practical recommendations

To summarise, disk access is one of the slowest parts of an OghmaNano simulation, especially on systems where storage has been configured for convenience rather than performance. The following guidelines can help you get the best out of your hardware:

• Run simulations on a local SSD or NVMe drive. Avoid network home directories for active work where possible.
• Avoid running simulations directly from cloud-synced folders such as OneDrive or Dropbox. Use a local working directory instead.
• Copy results after the fact to your home directory or cloud storage for backup and sharing, rather than simulating there.
• Check antivirus settings if simulations are unexpectedly slow, and consider excluding your simulation folder from real-time scanning (subject to local policies).
• Keep an eye on disk activity. If the CPU is idle but the disk is very busy, I/O is probably your main bottleneck.

Taking a few minutes to ensure simulations are running from the right location can easily save hours of compute time over the lifetime of a project, especially for large parameter sweeps, 2D/3D drift-diffusion runs, or heavy optical simulations.