Forum: CAT Tools Technical Help
Topic: Using a RAM disk to make CAT tools faster
Poster: DZiW
Post title: 12-18%
Samuel, while I don't work with Trados and big TMs, I don't think some 200MB is too big or unusual nowadays. With DDR3+ being more than two times faster than SDD and more than six times faster than HDD, your findings of 12-18% increase (if there's no HDD/SSD caching or swapping) indicate the bottleneck is not the media only, but rather CPU or even a non-optimal algorithm using one core--or limited in a way.
Unfortunately, it's quite common when a client hires students from developing countries to work under a project very cheap, so they often lazily rewrite optimal code in Java or C# forgetting about many bad things. For example, how about a Cpp programmer who doesn't care that boxed objects and collections are doubled at reaching a certain size, what immediately requires X + 2X -- triple amount of memory? And bruteforce checking instead of using hashes and indexes or other approaches? A cheap victory.
For what it's worth, as a part of the experiment and providing it's feasible, I'd like to see the RAMcahing in such a scenario too.
Topic: Using a RAM disk to make CAT tools faster
Poster: DZiW
Post title: 12-18%
Samuel, while I don't work with Trados and big TMs, I don't think some 200MB is too big or unusual nowadays. With DDR3+ being more than two times faster than SDD and more than six times faster than HDD, your findings of 12-18% increase (if there's no HDD/SSD caching or swapping) indicate the bottleneck is not the media only, but rather CPU or even a non-optimal algorithm using one core--or limited in a way.
Unfortunately, it's quite common when a client hires students from developing countries to work under a project very cheap, so they often lazily rewrite optimal code in Java or C# forgetting about many bad things. For example, how about a Cpp programmer who doesn't care that boxed objects and collections are doubled at reaching a certain size, what immediately requires X + 2X -- triple amount of memory? And bruteforce checking instead of using hashes and indexes or other approaches? A cheap victory.
For what it's worth, as a part of the experiment and providing it's feasible, I'd like to see the RAMcahing in such a scenario too.