Hi Lizz, comments in line On Wed, Nov 25, 2009 at 5:08 AM, laura pena <lizzpenaorclgrp@xxxxxxxxx>wrote: > Hello, > > We have been running linux 10g redhat for some 2 years now. Looking to port > to 11g. > Our linux systems run well, I hear windows has a new "architecture" and is > faster with 11g. > > There isn't anything fundamentally new about the Oracle on Windows architecture since the Windows NT initial release. I'd hope that 11g was faster than 10g on all platforms - though I still rather giggle at the whitepaper on upgrading to 11g from 9i that showed that if you did a whole bunch of tuning (and used extra cost packs) you could get 11g faster than 9i by about 10%. You don't say what version of Linux you are using (actually which kernel line 2.4 or 2.6 probably more important) but when I tested back in the 2.4 kernel days against Win2000 (dates the test somewhat) then Windows had a performance advantage of between 5 and 15% on TPC-C style benchmarks on the same hardware. This was mostly due to the superior I/O out of NTFS as compared to ext3. The big limitation for Windows is the thread based model (a good choice in the timeframe it was made by the way) which limits all memory to the limits imposed by the processor, on 32bit Windows this means the whole db SGA and PGA is limited to either 1.7 or 2.7gb (on Linux its just the SGA that is so limited) without resorting to somewhat nasty hacks for the buffer cache only. This has a double whammy since each new process eats up 1mb of that limit (alterable with orastack) so on systems with a few hundred dedicated server connections then you can hit memory limits - ORA-04030 - quite quickly. This problem basically goes away (until circa 2020 probably) with 64bit Oracle on 64bit Windows, but I wouldn't personally use that until at least Windows 2003 SP2 and Oracle 10.2. so if you are migrating then probably 64bit 11g on 64bt Server 2008 is a perfectly reasonable alternative from a technical point of view. Incidentally there is a new shared memory architecture with 11g on Linux. > Any known issues that we should be aware of if we were to port to Oracle > windows 11g? > > A comparison on this would be good if anyone has experience. > > Years ago the SGA was limited on windows and you can not use named pipes > like you can on linux. > depends on what you mean by 'use named pipes' You can't create named pipes from the os using the equivalent of mknod, but is that really such a common technique now? I know I used to use it for compressing exports in the days of low disk space and 2gb file limits, but the latter never really applied to windows and the former is less of an issue these days (though I was at a client yesterday whose super storage apparently costs them GBP7 per Gb so they are somewhat careful with allocating it!) > Windows was slower and you had to reboot the OS weekly (Oracle9i, can't > remember what windows server). > Sounds like a not so well managed windows box - or one with a lot of extraneous stuff on it. The windows boxes I see now tend only to be restarted for security patch application, now admittedly that happens a lot more than the Linux boxes I see, but that's at least partly because Linux admins never seem to patch o/s security holes. > But now I am not sure how windows scales compared to linux on 11g. > > I'm not aware of any current tests. There is *some* merit in looking at the TPC-C results for 2-4 processor Intel boxes, I'd expect them to show similar scalability for similar hardware on those two operating systems. > Any thoughts, experience on this issue would be greatly appreciated. > > Thanks, > Lizz > > as Matt says, by an large its people that compromise systems and not hardware or software, therefore if you and your admins are happier, more skilled etc etc in Linux/*nix then a move to Win32 would be adding risk, if the Oracle db is the only Linux server in the business then maybe a move would be worth considering. This sort of consideration should almost always outweigh the technology. -- Niall Litchfield Oracle DBA http://www.orawin.info