Re: Swap Space
- From: Andrew Kerber <andrew.kerber@xxxxxxxxx>
- To: dedba@xxxxxxxxxx
- Date: Thu, 25 Mar 2010 21:32:44 -0500
The swap space is a problem if the system is using eager swap, that is if
the OS marks swap as allocated when the RAM as allocated. When the system
uses eager swap, and the swap space isnt set per the recommendation, and if
your database uses a significant amount of memory, you are likely to run
into problems that look like problems with insufficient memory, even when
it is actually a problem with insufficient swap.
On Thu, Mar 25, 2010 at 9:13 PM, De DBA <dedba@xxxxxxxxxx> wrote:
> In my experience this should read: "... will likely crash the database.".
> The Linux kernel includes a process called "oom-killer", which in an
> Out-Of-Memory (OOM) situation will try to identify and forcefully kill idle
> memory hogs using some advanced algorithms.
>
> Unfortunately with Oracle processes, the most important such as SMON in the
> database and OPMN in the oracle app server may seem perfect candidates for
> this as they tend to be idle most of the time and allocate a lot of memory.
> Killing these of course is not a good idea, but it does free up memory. It
> may also kill other vital but usually near-idle processes, causing the OS to
> become unresponsive.
>
> The oom-killer logs its actions to syslogd, so you can always find a trail
> showing why the database crashed (assuming that /var is not full and you do
> log kernel messages).
>
> I am not sure that kswapd pushes up the average load by itself, as this
> reflects the average number of runnable processes, but I am not a kernel
> expert by any measure so I may well be wrong here. Incidentally, a load of
> 40 doesn't strike me as awfully high on 8 (16?) cores. It works out to
> 2.5-5/core. I've seen dual core machines work quite happily, if not
> extremely slowly, with loads up to 10-15 per core. At any rate, the number
> of runnable processes is not a reliable measure for memory usage.
>
> Cheers,
> Tony
>
>
> On 26/03/10 6:26 AM, Martin Bach wrote:
>
>> Hi Bill,
>>
>> just my 0.02 worth. The amount of swap the installer requires is a bit mad
>> with lots of memory like Howard already pointed out. The amount of swap
>> you need depends on your application, bear in mind that if the system runs
>> out of memory and swap it will crash. Before that it will become quite
>> unusable since the kswapd daemon(s) will agressively try to free memory.
>> This process can and will push the load average very high, and if no
>> substantial amount of memory can be freed causing kswapd to sleep again
>> your system is likely to become completely unresponsive to the point where
>> it will have to be rebooted. Just happened to me today-the last stats top
>> reported was a load average of 40 with a 8 dual core opteron box, but I
>> expect it went much higher than that.
>>
>> So if your application is gentle on memory then you only need a little
>> amount of swap (just for that warm fuzzy feeling), but if you know that
>> the box is going to be hit hard you might want to add some extra swap
>> space for peace of mind. Once you did that, tune the application to use
>> less memory-there are plenty of articles out there helping you with that
>> task.
>>
>> Hope that helps,
>>
>> Martin
>>
>> No the estimates by the installer get a bit mad with a large amount of
>>> memory
>>>
>>> On 25/03/2010, Bill Zakrzewski<bill@xxxxxxxxxxxx> wrote:
>>>
>>>> I am installing Oracle 10.2.0.4.0 on a linux server running RH 5.x with
>>>>
>>> ...
>>
>>> Thanks in advance,
>>>> Bill--
>>>> //www.freelists.org/webpage/oracle-l
>>>>
>>>> --
>> Martin Bach
>> OCM 10g
>> http://martincarstenbach.wordpress.com
>>
>> --
>> //www.freelists.org/webpage/oracle-l
>>
>>
>>
>>
>> --
> //www.freelists.org/webpage/oracle-l
>
>
>
--
Andrew W. Kerber
'If at first you dont succeed, dont take up skydiving.'
Other related posts:
