Re: Informix SE on "Raw" filesystems - bunk?

In article <467bk2$376@news.informix.com> aland@informix.com (Colonel Panic) writes: >In article <464j5f$lht@numbers.wetware.com>, >Scott Ellard <cse@wetware.com> wrote: >>In article <461kgo$jje@news.informix.com> aland@informix.com (Colonel Panic) writes: >>>>Just been going thru old partially-read email and discovered a "correction" >>>>to the FAQ from a fellow who claims you can run SE on "raw" file-systems. >>>> >>>>Sounds like bunk. Looks like bunk. Smells like bunk. What can it be? >>> >>>It's bunk. >> >>Alan, My friend, it can be done. Scott > >Modern flock()/fcntl() SE database can be run on raw devices? We're >all ears. Wow. I really should have qualified my answer although I made no reference to "modern" SE databases. Please accept it more in the realm of "is it possible in the absolute sense" as opposed to being something of general use. What I do remember is that way back in the Informix Turbo days when the consulting group was first dealing with raw partitions I played around in SE with moving a .dat file to one raw partition and the .idx file to another raw partition. It did work and did give improved response times when doing sequential scans of that particular table. The issue Alan quite correctly raises regarding flock()/fcntl() wasn't relevant since I was on a Sun which at that time had no support for flock()/fcntl(). SE used .lok files. Just for fun, I re-created that experiment today on SunOS 4.1.3 with Informix version 5.00.UC1 and I was able to move the .dat files but not the .idx for the reason Alan mentions. (the fcntl() call fails on the .idx file) It's a stretch, but I imagine there could be a table in SE that could benefit from being on a raw partition. However on a machine to machine basis there are other issues to consider. Some require reads/writes to/from raw devices to use buffers that are aligned on special boundaries due to DMA requirements. My apologies to the group and Alan for not being more precise in my original response. Scott