[mrtg] Re: linux performance

[Pavel Ruzicka]

Hello,

I am polling about 17000 targets.
I have mrtg-2.14.3 with RRDTool and 14all.cgi
System is older linux with kernel 2.6.18, dual Xeon 2.40GHz
with 3 GBytes of RAM.

There are LSI Logic megaraid with two 15000rpm disks in a mirror.
When everything is OK, MRTG runs for about 45 seconds.
All RRD files are probably cached by kernel in a RAM. Next kernel
writes caches to a disk for about 30 seconds on my machine. Now is
machine totally dead. I don't know what with this situation.

I have few things optimized on this server:

1. Very fast harddisks and fast controller. I have LSI Megaraid U320
and two SCSI 15000RPM 18GB harddisks in a mirror.
I need faster disks:-)

2. Enough of RAM. I have 3GB, but you need probably minimum 4GBytes of RAM for 
25000 targets. If your system use swap, it is probably signalizing, that 
system doesn't have enough of RAM for caches. My system use about 4k
of swap.

3. Optimized filesystem for many files. I have turned on feature "dir_index"
on ext3 partition with data files. Next I have mounted partition with option
"noatime". Maybe ext2 or other filesystem will be faster, but I don't know...

4. I have four virtual processors and I call MRTG from cron with 20 forks.
System balances load between all processors. But CPU power is not critical.

5. I have a little changed SNMP parameters with --snmp-options=:::2::2
For details look at manual. There are still reserves with tuning for me...

6. Good network card. I have onboard Intel Gigabit ethernet card connected
100Mbps-FD. Intel or 3Com will be probably good. Realtek probably not.

7. New versions of MRTG have "dead host detection". It means, that your
"dead" device with hundred interfaces doesn't block mrtg from running.
Unfortunately this feature doesn't work for "very busy" devices.
One overloaded router can block completely MRTG.
When this situation occurs, that there are some kind of problem with
locks and new MRTG starts while old still running.
I don't know if this is problem of my system or some interforks lock
problem. When I try run second MRTG manually, that locking works ;-)

Have somebody tips for better multitasking, when kernel flushes caches?
I tried tune kernel preempting and schedulers, but without visible change.

> I may be trying to do more than MRTG was ever designed to do but right
> now I am polling about 25,000 interfaces with a Dell 2850 running Red

Best regards,

Pavel Ruzicka

--
Unsubscribe mailto:mrtg-request at list.ee.ethz.ch?subject=unsubscribe
Archive     http://lists.ee.ethz.ch/mrtg
FAQ         http://faq.mrtg.org    Homepage     http://www.mrtg.org
WebAdmin    http://lists.ee.ethz.ch/lsg2.cgi