* Arno Wagner <arno@xxxxxxxxxxx> wrote: > On Wed, Sep 17, 2008 at 10:25:47AM +0200, Michael Gebetsroither wrote: >> 30MB/s? >> You know that even a single moderm SATA disk does 100MB/s streaming? > > Not really. Most reach close to 100MB/s at the start and go down to > something like 50MB/s at the end. This was a very modern notebook > disk (low power server) towards the end. Except those lovely ssd's from which you get about 170-200MB/s from one disk. >> The _real_ problem behind all these is much bigger. For the majority of >> the users the single-threaded performance will stagnate (or even >> decrease) in the next years > > So, they bought not what was actually fastest. For most, it will > still be more than fast enough. No, not that they bought a slow PC. A new quad-core is quite a bit slower in single threaded performance as a high-end dual-core. Even if the quad-core has much greater multi-threaded performance, single threaded workloads are not their friends. Intel just released a 6x core and the 8 cores are on the way to the mainstream. Thats what i meant with single threaded performance is about to drop quite noticeable in the next years. Just take the sun T1/T2 as example. Even if they have hw crypto engines for every core, you have to feed them from multiple threads to saturate a Gbit link. >> > Incidentially, I believe your request has about zero prospect >> > of being sucessful. It is a lot of effort for basically >> > very few people having any gain. I would not do it unless fully >> > paid, but thet _wpuld_ be expensive. >> >> As the fix with the current situation is quite simple (put the raid on >> top of dm-crypt) it's not the biggest problem, yes. > > Well, if it is RAID, hat is a good solution. However for > single disks that does not work. I originally thought > this was about an encrypted fast flash disk. For really large files, e.g vides, it's possibel to make a few partitions with dm-crypt and combine them into a raid0 with large chunk size (to keep the seeking low). This setup is quite usable for streaming workloads but is unusable for anything other. > One the other side, the problem is not really an issue, > even though some people seem to feel strongly about it. Not a big issue yet, ack. I just wanted to note that the situation will get a lot worse in the next years. cu, michael -- It's already too late! --------------------------------------------------------------------- dm-crypt mailing list - http://www.saout.de/misc/dm-crypt/ To unsubscribe, e-mail: dm-crypt-unsubscribe@xxxxxxxx For additional commands, e-mail: dm-crypt-help@xxxxxxxx
