Hi guys
In my environments, usually I configure OS's memory treshold just on swap. The
reason is that, in my knowledege, Linux (using kernel 2.6) as well Windows, pre-
allocate most of the memory and, in my experience, only swapping systems could
have problems. I used the conditional because, rarely, I even had some machines
running oracle with 100% of physical ram used + some swap without any problems
(no slow queries or degradation). But this could be just a special case
(oracle managing memory in aggressive way?) because generally, systems using to
much swap don't perform very well.
But...the fact that I always did in this way, doesn't mean that my reasoning
is good
Looking in the memory rule section of WATO, there are lot of items that should
be checked
Levels for RAM
Levels for Swap
Levels for Total virtual memory
Upper levels for Total Data in relation to RAM
Upper levels for Shared memory
Upper levels for Page tables
Upper levels for Disk Writeback
Upper levels for Committed memory
Lower levels for Commit Limit
Lower levels for Largest Free VMalloc Chunk
Handle Hardware Corrupted Error
The question is:
what is your opinion? what threshold do you usually apply to memory for both
Linux and Windows servers?
Guess it depends on your version - I don’t see all the options you are showing there but I’m running RAW.
For my instances, I usually ignore the used memory since Windows )in this example) is made to use as much of it as possible when you’re using SQL and Exchange, etc. So monitoring it seems pretty worthless since it’s going to use it all unless you put in a manual cap.
Swap File? That’s the one for me. When I see windows servers get to thresholds via the regular agent settings that go above 90%, the server will feel slower to the users. Then, we get the alert and will usually schedule a reboot unless we can remedy it in another way.
It doesn’t happen a lot, but that’s the one that people will report performance issues on when they have low resources to begin with.
That’s for our environments, anyway.
···
–
Brian Binder
On February 14, 2017 at 8:38:06 AM, mlist@libero.it (mlist@libero.it) wrote:
Hi guys
In my environments, usually I configure OS’s memory treshold just on swap. The
reason is that, in my knowledege, Linux (using kernel 2.6) as well Windows, pre-
allocate most of the memory and, in my experience, only swapping systems could
have problems. I used the conditional because, rarely, I even had some machines
running oracle with 100% of physical ram used + some swap without any problems
(no slow queries or degradation). But this could be just a special case
(oracle managing memory in aggressive way?) because generally, systems using to
much swap don’t perform very well.
But…the fact that I always did in this way, doesn’t mean that my reasoning
is good
Looking in the memory rule section of WATO, there are lot of items that should
be checked
Levels for RAM
Levels for Swap
Levels for Total virtual memory
Upper levels for Total Data in relation to RAM
Upper levels for Shared memory
Upper levels for Page tables
Upper levels for Disk Writeback
Upper levels for Committed memory
Lower levels for Commit Limit
Lower levels for Largest Free VMalloc Chunk
Handle Hardware Corrupted Error
The question is:
what is your opinion? what threshold do you usually apply to memory for both
Linux and Windows servers?
For SQL and the like, we make the application folks set max memory settings. Otherwise you won’t know when the OS goes sideways. If it has 16gb of RAM, we recommend setting it at 12gb or 13gb at the most. For those footprints, we like the OS to have at least 1 GB free before it starts alerting.
We ignore pagefile everywhere since most systems shouldn’t be paging anyways. Haven’t monitored pagefiles in over 10 years.
If windows servers start to slow down, it’s probably because they are misconfigured. Google “russinovich pagefile size” for a detailed blog post on why.
Guess it depends on your version - I don’t see all the options you are showing there but I’m running RAW.
For my instances, I usually ignore the used memory since Windows )in this example) is made to use as much of it as possible when you’re using SQL and Exchange, etc. So monitoring it seems pretty worthless since it’s going to use it all unless you put in a manual cap.
Swap File? That’s the one for me. When I see windows servers get to thresholds via the regular agent settings that go above 90%, the server will feel slower to the users. Then, we get the alert and will usually schedule a reboot unless we can remedy it in another way.
It doesn’t happen a lot, but that’s the one that people will report performance issues on when they have low resources to begin with.
That’s for our environments, anyway.
–
Brian Binder
On February 14, 2017 at 8:38:06 AM, mlist@libero.it (mlist@libero.it) wrote:
Hi guys
In my environments, usually I configure OS’s memory treshold just on swap. The
reason is that, in my knowledege, Linux (using kernel 2.6) as well Windows, pre-
allocate most of the memory and, in my experience, only swapping systems could
have problems. I used the conditional because, rarely, I even had some machines
running oracle with 100% of physical ram used + some swap without any problems
(no slow queries or degradation). But this could be just a special case
(oracle managing memory in aggressive way?) because generally, systems using to
much swap don’t perform very well.
But…the fact that I always did in this way, doesn’t mean that my reasoning
is good
Looking in the memory rule section of WATO, there are lot of items that should
be checked
Levels for RAM
Levels for Swap
Levels for Total virtual memory
Upper levels for Total Data in relation to RAM
Upper levels for Shared memory
Upper levels for Page tables
Upper levels for Disk Writeback
Upper levels for Committed memory
Lower levels for Commit Limit
Lower levels for Largest Free VMalloc Chunk
Handle Hardware Corrupted Error
The question is:
what is your opinion? what threshold do you usually apply to memory for both
Linux and Windows servers?
