does the installed memory is sufficient ? -- plz help

[Maaz]

the output of /proc/meminfo attached.
this machine is under the typical load.
does the physical installed memory is unsufficient/sufficient ?

Regards
Maaz

[Alexander Chuzhoy]

Although the best way to test the load o f the machine is to use the utilities (like vmstat, sar, etc), from the attached file it seems like the swap isn't being used at all, thus the amount of the installed memory is sufficient for the curent load.

[Ivan Ferreira]

I agree with Alexander, but to truly know if you need more memory, you should check the vmstat output if you have page ins AND page outs (sar can provide that information also). If so, then you need more memory.

Also you need to consider the response time from the user perspective. For example, the system may not page, but users compains about performance. This could mean that you should tune your application to use more memory, if you do that, you may start paging.

[Steven E. Protter]

Shalom Maaz,

The criteria for sufficient memory is usually via tests.

If the system is paging to disk, not just reserving swap, then get more memory in the system.

The vmstat/sar utils should be run while the system is under heavy use.

One thing you can do is a memory use estimate.

Check your oracle SGA sizes and see what the total adds up too. Similar techniques can be uses for other software. Example, software ag requires 20 MB of server memory for every remote user.

SEP

[Maaz]

thanks everybody for help.

This machine is the fileserver(samba, and nfs).
vmstat 3, was executed when the machine was under the max load.
the output of the 'vmstat 3' is attached.

Ivan
>the system may not page, but users compains about performance
we have recieved mix output from the users, some says performance is not sufficient(it should be faster) some says it sometimes slow, but normally work reasonable etc etc.

please advice.

[Ivan Ferreira]

From the output, your performance problem is not relaten to memory, is related to I/O, and this could be network or disk, as you have high "I/O wait" (wa column for CPU statistics).

The "wa" column means that you have CPU that is not doin any processing because is waiting for I/O.

You need to check:

- Your disk subsystem and filesystem performance (iostat/hdparm/sar/bonnie).
- Your network statistics and performance (sar/collectl).

[Maaz]

Thanks Ivan for help.
yes, as this is a fileserver(samba/nfs server) it is obvious that I/O operations are too much all the time.

1, Should I use the SAS harddrives instead of SATA/SCSI ?

2, so which value(number) will be consider/suppose as "High" or "too much" for "wa" column ? I mean what should be the max value for "wa" column of a fileserver for optimum performance ?

3, whats the difference b/w "Buffers" and "Cached" in the output of /proc/meminfo ? I know that Cache is good, because this is the memory Linux kernel occupy in adv for optimum performance, but what is Buffer ?

Regards

[Maaz]

4, which columns and what values shows that processor is fast enough for optimum performance of a (a)fileserver (b)Squid Proxy Server, and (c)Oracle 10g Server


5, what will be the max value for "r" and "b" column for Procs statistics of a (a)fileserver (b)Squid Proxy Server, and (c)Oracle 10g Server ?

Regards

[Ivan Ferreira · Accepted solution]

>>> 1, Should I use the SAS harddrives instead of SATA/SCSI ?

Maybe, SAS provides 300 MB/s, SATA II should also provide 300 MB/s, and SCSI 320/360 MB/s.

In this case, the problem should not be the disk, but the disk configuration.

Consider using disk striping or RAID 0+1 for performance.

Split your data in several disks.

Consider the usage of multiple ethernet cards.

>>> 2, so which value(number) will be consider/suppose as "High" or "too much" for "wa" column ? I mean what should be the max value for "wa" column of a fileserver for optimum performance ?

To me, the "wa" should be close to 0, but this depends of the performance "percibed" and the host hardware/configuration.

For example, on my systems, for a database server (large servers), "wa" of 10 is reached when backup is performed. But remember, this depends of the hardware and the disk subsystem.

>>> 3, whats the difference b/w "Buffers" and "Cached" in the output of /proc/meminfo ? I know that Cache is good, because this is the memory Linux kernel occupy in adv for optimum performance, but what is Buffer ?

There are a few "buffers" in the kernel infraestructure. Search for "buffer" in the kernel glosarry:

http://people.nl.linux.org/ftp/pub/anoncvs/kernelnewbies/glossary/

>>> 4, which columns and what values shows that processor is fast enough for optimum performance of a (a)fileserver (b)Squid Proxy Server, and (c)Oracle 10g Server

Talking exclusivelly about processors, you should have always some "idle" processor. You should have high values in user (us) and lower in system (sy) processor statistics, this means that the CPU is used for user applications and not for system operation (as paging).

But you need to consider the whole subsystem, that is, idle processors, no pagig, I/O wait low.

>>> 5, what will be the max value for "r" and "b" column for Procs statistics of a (a)fileserver (b)Squid Proxy Server, and (c)Oracle 10g Server ?

You can't say, this really depends of each system. Consider other statistics as the "load average", paging and I/O wait.