Is it possible to trigger a bogus transaction say, an hour before your first transaction is usually executed?
We had a similar problem with CICS back in the '90s. The transaction did a call (a no no back in CICS 1.7) to an external routine which did a lot of paging and a lot of I/O to non-cached controllers. The first time thru, the transaction would abend (AICA...runaway task timer). After storage was loaded, it would run in a few seconds. The cheapest/easiest solution was to trigger the transaction a couple times before users got on. Is that possible with your application? Tom Duerbusch THD Consulting >>> Mark Wheeler <mwheele...@hotmail.com> 3/17/2011 11:55 AM >>> Bob, No, the question was asked previously, but I chose to ignore it. For one, because it would take way too long to explain adequately, and also that the thread would quickly expand exponentially. Quick answer: we have an app that sees elongated response times on the first transaction of the day. We have traced it back to several thousand synchronous pageins (because stuff got paged out overnight, and our page volumes aren't infinitely fast). All subsequent transactions run sub-second. I have no idea which pages are involved, but it was suggested that since they were synchronous pageins, it may involve the kernel. A POSSIBLE solution that crossed my mind would be to lock kernel pages in storage and see if that solved the problem. All I needed to conduct that little experiment was to know where the kernel lived in storage. Again, I know most everyone who reads this will have the same obvious questions and suggestions, and I appreciate that. Alas, right now there isn't enough time or bandwidth to explain the situation in sufficient detail so as to prevent this thread from blowing up in a hundred different directions. The suggestion to look at /proc/iomem answered my immediate question. As necessary, I'll toss further questions out to the list. Thanks all! Mark Wheeler UnitedHealth Group > Date: Thu, 17 Mar 2011 11:30:27 -0600 > From: nix.rob...@mayo.edu > Subject: Re: Where is kernel loaded in memory? > To: LINUX-390@VM.MARIST.EDU > > An obvious question that no one has bothered to ask as yet: > > What is the problem you're trying to solve with this? Or, why do you want to > know where the kernel loads, and what will you gain from it? > > Too many times, users or other people (programmers, other sysadmins, ...) > come to us with a solution in need of a piece or part, and we never hear the > larger question or problem, to which there may be a much simpler answer. > > The query may be a simple one, the need may be educational. Or it may be a > cog in a larger, complex solution to a problem that some, or many of us have > already solved in some other way which does not involve walking through the > kernel's memory. > > It's just a thought, but Mark -- What's your original problem or task? > > -- > Robert P. Nix Mayo Foundation .~. > RO-OC-1-18 200 First Street SW /V\ > 507-284-0844 Rochester, MN 55905 /( )\ > ----- ^^-^^ > "In theory, theory and practice are the same, but > in practice, theory and practice are different." > > > > On 3/17/11 5:59 AM, "Richard Troth" <vmcow...@gmail.com> wrote: > > > Originally, the kernel loaded at "real" addr 64k. That is the default for > > Linux on most platforms. But you could change that, and for 1M alignment, > > some do so on S/390. > > > > Going with mapped memory, it sounds like absolute zero is the virtual pref > > for kernel space. Cool. Easily handled in all virt mem platforms. > > > > -- R; <>< > > ---------------------------------------------------------------------- > For LINUX-390 subscribe / signoff / archive access instructions, > send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit > http://www.marist.edu/htbin/wlvindex?LINUX-390 > ---------------------------------------------------------------------- > For more information on Linux on System z, visit > http://wiki.linuxvm.org/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
