Hi Julien, Julien BLACHE wrote: > abel deuring <[email protected]> wrote: > > Hi, > >>> I think we can even get rid of the old SG interface entirely. SG_IO >>> exists in Linux 2.4 too, so it should be safe. >> I must admit that I am one of these persons who cannot throw away >> anything... But you are right: Doug Gilbert wrote the SG3 interface in >> the late 90ies (IIRC, the first versions were available even for 2.2.x >> kernels), so keeping the old interface would be useful for very few >> people, if any. > > According to <http://sg.torque.net/sg/> SG3 was first released in a > stable kernel with Linux 2.4.0 in 2001. It's unclear to me whether > it's been available in Linux 2.2 before that. (are there really people > here still running SANE on a 2.2 kernel ?)
It seems that my memory it not very well functioning ;) Perhaps there were some some beta versions available earlier, but never mind. > >>> The latency here seems low enough, and this machine is an SGI Indigo2 >>> R4400SC 200 MHz w/256 MB RAM and an asthmatic SCSI controller (under >>> Linux, at least, because we don't have the specs ...) >> The effects of longer latency times seem to depend heavily on the >> scanner models. The Sharp JX250, a moderately fast model, probably with >> only a tiny internal buffer, seems to suffer more than many other devices... > > It'd be nice to gather some feedback on my patch with a couple of > different scanners, that'd set the record straight as far as latency > goes :) It could be that the JX250 is much more sensistive to latency problems than other scanners... I'll test your patch during the next days (sorry, I can't make at present a more precise and honest promise...). But I already made some test scans with sane-backends 1.0.18 and a modified sharp.conf file that disables queueing on the SG level: No performance problems with a 1.6 GHZ Athlon mainboard and a SCSI controller for the sym53c8xx driver. (I don't have a slower system available...) Abel
