Not GP but one disadvantage of updating one huge file is it's harder to do efficient incremental backups. Theoretically it can still be done if your backup software supports e.g. content-defined chunking (there was a recent HN thread about Google's rsync-with-fastcdc tool). If you choose to store your assets as separate files instead though, you can trivially have incremental backups using off-the-shelf software like plain old rsync [1].
If you're using 16 PCI 4.0 lanes you max out at 32GB/s, although commercial drives tends to have much lower throughput than that maximum (~7.5GB/s for a good NVMe drive). Cat6a ethernet tops out at 10 gigabits per second, but plenty of earlier versions have lower caps e.g. 1 gigabit. My guess is you'll most likely be limited by either disk or network hardware before needing CPU parallelism, if all you're doing is copying bytes from one to the other.
Not GP but one disadvantage of updating one huge file is it's harder to do efficient incremental backups. Theoretically it can still be done if your backup software supports e.g. content-defined chunking (there was a recent HN thread about Google's rsync-with-fastcdc tool). If you choose to store your assets as separate files instead though, you can trivially have incremental backups using off-the-shelf software like plain old rsync [1].
[1]: https://www.cyberciti.biz/faq/linux-unix-apple-osx-bsd-rsync...
> there was a recent HN thread about Google's rsync-with-fastcdc tool
Was this the tool & thread you mean? https://news.ycombinator.com/item?id=34303497?
Yeah that's the one!
1 reply →
Wow, that is actually an amazing performance curiousity adding parallelism to the mix. I guess this would depend on the M.2 spec?
If you're using 16 PCI 4.0 lanes you max out at 32GB/s, although commercial drives tends to have much lower throughput than that maximum (~7.5GB/s for a good NVMe drive). Cat6a ethernet tops out at 10 gigabits per second, but plenty of earlier versions have lower caps e.g. 1 gigabit. My guess is you'll most likely be limited by either disk or network hardware before needing CPU parallelism, if all you're doing is copying bytes from one to the other.
The other being a network socket in this case? But that socket might be two servers over? Meh, ideally they've optimized that as well.
So absolutely it is a network problem which means custom fiber?
1 reply →