Re: Informix Online: Shared Memory Problem (Linux)
Posted in 1995
In article <3tqmvc$ntq$1@mhafm.production.compuserve.com>, Paul Redman <73764.1657@CompuServe.COM> wrote: >I had a similar problem in Online version 5 for SCO and the >README file did not help for the size of SHMAX environment >variable. > >You have to increase this tenfold (which sounds daft, but is >absolutely correct for SCO) in your kernel and suddenly the >engine starts to work. > >Could be the same for Linux - try it and see. > >Paul Redman The problem is a bug in the "ipc.c" portion of the iBCS module. Get the April '95 version of iBCS-1.2x and get the patch file (something-or-other).patch1 from tsx-11.mit.edu. You'll probably have to apply the patch by hand to the ipc.c source code file because the fellow who created the patch used a hard subdirectory path that doesn't exist in the un-tarred contents of the iBCS source tree and I don't think the line numbers in the patch file line up exactly with the original ipc.c file, but if you look in the neighborhood of the line numbers stated in the patch1 file, you'll see the lines of code that need to be altered (assuming you know some "c" programming) to implement the bugfix. Use at least kernel version 1.2.10, preferably 1.2.11 and you'll be pleasantly surprised with the results (after you install the compiled iBCS module). IBCS *does* work and Online 5 can run, even using the default std tbconfig values, even on an 8-meg ram Linux machine though, with what performance figures is left as exercise to the experimenter. You'd probably have to use cooked filespace chunks, I have no idea if raw character device diskspace would work under Linux or not. Also, you probably won't be able to make any "C" compiler 4gl or ESQL/C apps since the link phase would attempt to link your compiled-natively-for-Linux applications objects to compiled-natively-for-SCO Informix library stuff, and the linker would most certainly bitch about that. I'm pretty sure that the iBCS support only lets the kernel execute completed, compiled-for-SCO executable binary files but somewhere out there, someone will undoubtedly come up with an OCT (object code translator)linker that can create a Linux executable out of a heterogeneous (but compiled for x86 pc unix variants) collection of object modules & libs. (A Linux OCTing,THUNKing Linker/Translator? :-) :-) :-) I also don't know much about the internals of the RDS version of 4gl, but if it doesn't use gcc and Linux's ld, then it might work if it compiles to its p-code and links its objects and libraries with a proprietary compiler & linker, provided that the RDS's compiler, linker, and fglgo p-code runner all will run okay under iBCS. peace, Neal -- ===================================================================== = Neal Howard neal@metronet.com '91 XLH-1200 DoD #686 AMA NRA = = "Fuzzy Wuzzy was a bear, Fuzzy Wuzzy had no hair. If Fuzzy Wuzzy = = had no hair, then Fuzzy Wuzzy wasn't Fuzzy, wuz he ?!!!" = =====================================================================