Dylan, That's roughly right.
The main problem is that we're simply dumping C++ in-memory data structures to disk. This means that word alignment, byte ordering and data-type sizes need to be the same between machines. Assuming you have that, then it should work fine. We also don't do any data load validation (there is some stale CRC code, but I don't believe it's currently used). This means that data format problems usually only manifest as weird routing problems later on. It's something everyone would really like to fix, but hasn't been a high enough priority for anyone to spend time on. It's important that file-off-disk loading is fast, so any cross-platform file compatibility layer needs to be carefully designed to not add too much of a performance hit. Ideally, we can `mmap` to most of the data files, which could unlock low-memory/low-performance usage scenarios. daniel > On Aug 11, 2016, at 7:51 AM, Dylan Adams <dylan.ad...@gmail.com> wrote: > > Hi, > > I'm aware that there are some limits on the portability of OSRM files, > as documented on issues #2242 and #1685. I've also seen evidence of > workflows that extract/contract on one machine, and distribute the > results to other hosts > (https://lists.openstreetmap.org/pipermail/osrm-talk/2016-March/001141.html). > > To what extent are OSRM files portable? > > My assumption is that OSRM files are portable as long as the OSRM > binaries, dependencies, processor architecture, and OS are the same. > Is that correct? > > Thanks, > Dylan > > _______________________________________________ > OSRM-talk mailing list > OSRM-talk@openstreetmap.org > https://lists.openstreetmap.org/listinfo/osrm-talk _______________________________________________ OSRM-talk mailing list OSRM-talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/osrm-talk
