Problem with RCC for version 6.10

sasho_atanasow · Post by **sasho_atanasow** » Tue Sep 12, 2006 9:41 am

Hello, my problem is the following:
1. Machine on which is running Hostmonitor v 6.10 is Pentium 4 - 3.2GHz with 2 Gbytes of RAM.
2. Four machines Pentium 4 - 3GHz with 512 Mbytes of RAM, connected to the first with RCC.
3. 728 tests - 464 SNMP tests, 240 ping tests, 13 traffic monitor tests, 10 DNS tests, 1 URL test ; load - 36 tests/sec, test interval for each test - 20 seconds.
4. Gigabit network connection on the machine, running hostmonitor, and 100Mbit connection for the machines running RCC.
The problem is that RCC loses connectivity with the hostmonitor, and the machine, running hostmonitor tests often gets busy to 100 % cpu load.
Can anyone point me a solution for the problem

P.S. The problem is very clearly seen, when more than 10-12 tests fail. But it exists, when everything is ok also.

Thank you in advance.

KS-Soft · Post by **KS-Soft** » Tue Sep 12, 2006 10:58 am

May be too short timeout interval? What timeout have you specified?

and the machine, running hostmonitor tests often gets busy to 100 % cpu load.

Do you use ODBC logging? If yes, what driver do you use?

Regards
Alex

sasho_atanasow · Post by **sasho_atanasow** » Mon Sep 18, 2006 3:52 am

Timeout for RCC is 30 seconds. The status refresh rate for RCC is 10 times/minute.
We don't use ODBC logging.

KS-Soft Europe · Post by **KS-Soft Europe** » Mon Sep 18, 2006 4:57 am

Could you increase RCC timeout to 45 sec?
I think, you also should increase timeout, specified on hostMonitor > Options > RCI page.
Anyway, significant CPU load can be explained: 10 times per minute HostMonitor should encrypt 728 tests results for 4 RCC clients. 40 times per minute! It takes a lot of time... and CPU usage. Perhaps, we need some improvements here.

Could you decrease "Status refresh rate" prameter?
Quote from documentation:
=============
Define this parameter depending on the network speed, HostMonitor's load and number of operators that may work with HostMonitor simultaneously.
=============
http://www.ks-soft.net/hostmon.eng/rcc/index.htm

Regards,
Max

KS-Soft · Post by **KS-Soft** » Tue Sep 19, 2006 1:49 pm

Anyway, significant CPU load can be explained: 10 times per minute HostMonitor should encrypt 728 tests results for 4 RCC clients. 40 times per minute! It takes a lot of time... and CPU usage. Perhaps, we need some improvements here.

Actually HostMonitor does not send information about test items that were not changed (not tested). So, it sends information about 36*6 = 216 test items every 6 sec to each RCC. Multiply by 4 RCC = 864 items. Not that bad but it leads to significant load. Plus HostMonitor should perform the tests, record the logs, etc.
We can try to oprimize some code but it cannot be done today...

On the other hand, do you really need to perform each test every 20 sec? What about 1 or 2 min? It will reduce system load in 3-6 times.

Regards
Alex

sasho_atanasow · Post by **sasho_atanasow** » Mon Sep 25, 2006 2:18 am

Actually HostMonitor does not send information about test items that were not changed (not tested). So, it sends information about 36*6 = 216 test items every 6 sec to each RCC. Multiply by 4 RCC = 864 items. Not that bad but it leads to significant load. Plus HostMonitor should perform the tests, record the logs, etc.
We can try to oprimize some code but it cannot be done today... On the other hand, do you really need to perform each test every 20 sec? What about 1 or 2 min? It will reduce system load in 3-6 times.

Regards
Alex

In fact, the goal is to achieve very rapid diagnostics, so we even want a test interval of 5-10 seconds. We'll try to optimize the tests, by removing eventual logically duplicated tests. But is there another way to optimize ?

sasho_atanasow · Post by **sasho_atanasow** » Thu Sep 28, 2006 12:12 am

We decided to not use RCC, and have single host monitor machine with a big TV screen attached to it.
There are 222 ping tests, 423 SNMP tests, 10 DNS tests, 8 Traffic monitor tests. Recently we expect a significant (20-30 %) increase in the number of tests. The test interval is 5 seconds for SNMP tests, and 2 seconds for the Ping tests. My question is what machine could bear the load? Will there be any use of putting Hostmonitor on SMP machine?

KS-Soft · Post by **KS-Soft** » Thu Sep 28, 2006 10:51 am

You want to perform about 200 tests per second

I am afraid that's impossible

HostMonitor will not start more than 128 tests per second. Theoretically we can increase this limit (just for you) but I am not sure system will work nice.
HostMonitor was not designed for such configuration - run all test items at the same time. Its better to use fixed set of threads for such task, while HostMonitor creates new thread for each test.

Regards
Alex

AntonyP · Post by **AntonyP** » Sat Sep 30, 2006 2:27 am

Just a friendly advice to sasho_atanasow

Have you got any idea how much this extreme interval of tests will increase the traffic in your internal network??

Honestly, i believe that the best way for you to check if a network device is alive or not is to set the interval of the test to 1min or more. If you believe that a device has a problem you'd better set an snmp trap instead of the 5-sec-snmpget.