Re: Informix under Linux - real crap !

I have tested only 9.30 UC2 on SuSE 8.1 (equals SLES 8). Definitly does not run. I think, we change to the newest 9.30 UC6 or 9.40 (to come out soon, according to informix support) , which is supposed to have those glibc 2.3 incompatibilities in RedHat 9.0 corrected. 2.4.19 - 64GB-SMP Kernel has several memory leaks an a very ugly sheduler for task switching. At high loads the machine hangs. Minimum for a usable Linux is (sorry to say that) 2.4.21 ... This kernel's behaviour feels like freebsd/netbsd/openbsd and runs very fine at high loads (NONSMP). What i am missing the most is KAIO. Without KAIO, Informix is not able to write into a chunk and to read from a chunk simultaneously. KERNEL ASYNC I/O is a minimum feature with SMP. Without, threads block each other, the speed improvement of SMP will be zero. A DUAL 66 MHz HP UX with old SCSI drives is only 3 times slower than a DUAL XEON (not used) 2.4 GHz with RAID (80 mbyte/sec I/O) I never expected this. I also think, that its time now to get that KAIO Libs integrated into Linux. I have RedHat 8.0 with glibc 2.3 kernel 2.4.19 running with KAIO. It was very complicated, hat to recompile many rpm packets. My experience with 30+ Users over ODBC on Informix 9.30 DUAL XEON without Multiprocessor Informix is quite strange. Informix default behaviour is, that if only one is running a longer task, all others will e blocked. This behaviour changes, when several paremeters in onconfig are changed. Why the heck is Informix not able to set correct default parameters for parallel use over ODBC ? Why does this /var/lib/libpthread.so.0 prevent Informix 9.30 UCx from running on SuSE and RedHat Linux ? There are several tuning guides out, any original informix howto out there ? regards, Guido Stepken Richard Kofler wrote: > Sandor wrote: > >>Hi Guido >>We have been successfully testing on a 2 physical processor SMP box with >>Hyper Threading turned on! (so cat /proc/cpuinfo shows 4) >>This can be verified nicely with top command or with >> >> onstat -g glo>> >>Individual virtual processors: >> vp pid class usercpu syscpu total >> 1 18406 cpu 35.77 0.67 36.44 >> 2 18407 adm 0.04 0.06 0.10 >> 3 18408 cpu 38.49 0.59 39.08 >> >>The OS was SuSE SLES 8 (2.4.19-64GB-SMP) 2.2.5-165 glibc (patched) and >>IBM Informix 9.40 >>There was some bad experience in the past with HT so this is why support >>was advising not to use it. >>Sandor >> >>Guido Stepken wrote: >> >>>>Please take a look to the article below how to install >>>>9.40 on different >>>>Linux distributions. As far as I know KAIO is not available in the >>> >>>official >>> >>>>sources of Linux 2.4.x kernels. But configuring multiprocessor in >>> >>>$ONCONFIG >>> >>>KAIO Patches are available since RedHat 7.1/7.2. >>> >>> >>>>should really work, e.g. >>>> >>>>MULTIPROCESSOR 1 # 0 for single-processor, 1 for>>>>multi-processor >>>>NUMCPUVPS 2 # Number of user (cpu) vps >>>>SINGLE_CPU_VP 0 # If non-zero, limit number of>>>>cpu vps to one >>> >>>No, does not really run. Only 1 CPU is really used. TOP >>>does not show >>>this . Hyperthreading is running fine with Linux Redhat 8.0 and >>>Informix. Why then should it be disabled ? >>>According to Informix support one should generally disable >>>multiprocessor support, because it does not run, it crashes. >>> >>> >>>>RedHat 9 >>>>-------- >>>>"Incorrectly built binary which accesses errno, h_errno or _res >>>>directly. Needs to be fixed." messages. Use 9.40.UC2 >>>>when available. >>> >>>Problem is, that 4GL/SQL Toolkit does not compile older >>>code. There is a >>> incompatibility between 9.30 and 9.40 addon's (not the >>>IDS itself), >>>which prevents us from going to 9.40. >>> >>>I really appreciate to know one distribution + patches, >>>which is able to >>>run Informix on SMP w or w/o hyperthreading. KAIO should >>>be a duty for >>>all database OS. Linux has it, why then doesn't Informix support it ? >>> >>>regards, Guido Stepken >> >>-- >>Direct access to this group with http://web2news.com >>http://web2news.com/?comp.databases.informix > > > I testing now SuSE 8.2 using glibc 2.3.2-9 and a sudo > script to overcome the mapped_base 'feature'. > > During 6 days of testing I did not have any problem > with Hyperthred Mode ON > > What problems should I expect? Anyone can give a hint? > > dic-K > -------------- > Richard Kofler > SOLID STATE > EDV-Dientleistungen > Vienna, Austria