[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On Thursday, 4 March 2021 19:27:09 GMT John PNZ wrote: > > My warped perspective was focused more on low-power hardware and core > temperatures. I've seen Bitlocker slap a 30° penalty on under-resourced > holiday video editing, for example. I assume to get 30% penalty you have to be deeply using swap? Curiously the relative penalty is getting worse with the arrival of SSDs, since with spinning rust the bottleneck was nearly always waiting for the disk, encrypting and decrypting added CPU overhead but few people are always maxed out on CPU. Of course the absolute penalty is less because CPUs and SSDs are both faster. The CPUs now have instructions to do the encryption, but we are all moving to SSDs, where seek time is gone, and the throughput is typically an order of magnitude faster (or more). This doesn't stop the Phoronix benchmarks showing that compiling Linux or compiling Apache is faster with full disk encryption enabled in Ubuntu on the nice DELL laptop they tried it on. I suspect if someone does a deep dive on this we'll find something bad happening in the compiler or linker IO pattern here, as full disk encryption really should have a penalty however small, and not be quicker even if only marginally. -- The Mailing List for the Devon & Cornwall LUG https://mailman.dcglug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq