Community
- Command Central
- MVP Program
- Monthly Mission
- Blogs
- Groups
- Events
- Media Vault
Products
- Observability
- Network Management
- Application Management
- IT Security
- IT Service Management
- System Management
- Database Management
Content Exchange
- SolarWinds Platform
- Server & Application Monitor
- Database Performance Analyzer
- Server Configuration Monitor
- Network Performance Monitor
- Network Configuration Manager
- SQL Sentry
- Web Help Desk
Free Tools & Trials
Store

Polling times and gaps in 5min graphs

I have set my polling and statistics collection perriods for very short perriods in an attempt to gather more detailed information.

I am polling nodes and interfaces every 60 seconds and have the statistics collection Interval set to 1 minute for Nodes & Interfaces. I have applied all of these settings to All, and even restarted my Orion server.

However, I still can only get a 10 minute resolution on my graphs. When I bring up a 5 minute graph, I show many gaps in the data. I even figured out that you can manually change a graphs link to show a 1 minute reolution on a graph to see exactly where the gaps are. Just change the part of the link where it says &SampleSize=5M to &SampleSize=1M. When I do that it shows that I have a data point every 6 to 7 minutes.

Interestingly enough, I went into the Performance monitor console and tried hitting the "poll" button on an interface lots of times and I then get minute by minute data on that interface because I forced polls.

Shouldn't I be able to get a higher resolution then 6-7 minutes with these settings???

Find more posts tagged with

Accepted answers

All comments

Network_Guru

You may need to tune your polling engine. Download and install the Polling Engine tuning tool. The default Maximum 8 polls per second is most likely too low for polling devices that frequently. You can also see the polling statistics under: File>Polling Status

You might be polling too frequently if you have any large switches or routers on your network. It probably takes longer than 60 seconds for the SNMP poll of all interfaces to reply. Unless your switches and routers are sitting around idling, doing nothing, it's not generally a good idea to poll that frequently. These devices were designed to switch/route packets, not respond to SNMP requests every 60 seconds. See how long it takes to discover all the interfaces on a fully loaded Cisco 6513....

-=Cheers=-
NG

iunderwo

The minimum polling time I've ever had any success with is 3 minutes. The resolution is better than 5, but not so much so that it will clobber the bejesus out of the device you're polling.

NGs spot on the polling tuner is also right on...that can make quite a difference.

// Ian Underwood - Service Management
// Level 3 Communications

rwbuckn

quote:Originally posted by Network_Guru

You may need to tune your polling engine. Download and install the Polling Engine tuning tool.

How much do you trust the tuner? I've been told by SW support not to use that tool. In fact when I do the math manually, I come out with a number of 69 polls per second, when the tuner shows 32 status polls per second and 57 collections per second. SW tells me to make the 2 numbers the same. What to believe?

Network_Guru

quote:Originally posted by rwbuckn

quote:Originally posted by Network_Guru

You may need to tune your polling engine. Download and install the Polling Engine tuning tool.

How much do you trust the tuner? I've been told by SW support not to use that tool. In fact when I do the math manually, I come out with a number of 69 polls per second, when the tuner shows 32 status polls per second and 57 collections per second. SW tells me to make the 2 numbers the same. What to believe?

It's a starting point at least. The tuner recommends 12/second for my network, and after testing over several months, I have found 24 polls/sec max for both numbers seems to work best in my environment, especially after the Orion server reboots. I was pointing out that the tool is available, and these parameters are available to finetune your installation.

-=Cheers=-
NG

pancamo

.. ...... ... .. ............ .......

We also have been having problems with gaps. (currently using v7.4) Every morning during the SW Database maintenance 2:15am we get about an hour gap on ALL devices where no CPU/interface snmp polls go through. We also have gaps on all the devices throughout the day at random times and random durations.

I have put a sniffer on the SW box, and watched for snmp polls going to the devices. SNMP packets are not leaving the SW box.

I have observed that the NetPerfMonServi service is running 100% during the day times, while normally it runs ~45% At 2:15am the disk get too busy (0% Disk Idle on the SW Database disk) and the NetPerfMonServi service goes from using ~45% CPU to 10% CPU.

I'm currently looking into moving the Database from the local raid 0 disks to a SAN disk, hopefully taking the load off the SW server. This should at least resolve the 2:15am hour gaps..

quote:Originally posted by ggardner1

I have set my polling and statistics collection perriods for very short perriods in an attempt to gather more detailed information.

I am polling nodes and interfaces every 60 seconds and have the statistics collection Interval set to 1 minute for Nodes & Interfaces. I have applied all of these settings to All, and even restarted my Orion server.

However, I still can only get a 10 minute resolution on my graphs. When I bring up a 5 minute graph, I show many gaps in the data. I even figured out that you can manually change a graphs link to show a 1 minute reolution on a graph to see exactly where the gaps are. Just change the part of the link where it says &SampleSize=5M to &SampleSize=1M. When I do that it shows that I have a data point every 6 to 7 minutes.

Interestingly enough, I went into the Performance monitor console and tried hitting the "poll" button on an interface lots of times and I then get minute by minute data on that interface because I forced polls.

Shouldn't I be able to get a higher resolution then 6-7 minutes with these settings???

Network_Guru

I believe the polling engine stops writing to the database during nightly maintenance.
Even compacting/re-building the table indexes stops the polling engine, since it can't write to the database during this time.

-=Cheers=-
NG

pancamo

that's not good... We loose an hour worth of data at 3am each morning...

We run a 24/hr shop here... Is there a way around this?

quote:Originally posted by Network_Guru

I believe the polling engine stops writing to the database during nightly maintenance.
Even compacting/re-building the table indexes stops the polling engine, since it can't write to the database during this time.

-=Cheers=-
NG

DTuloJr

I've changed my polling to 2 min for stat collection, but I'd rather have it only at 5. It seems when an SQL server on a system goes psycho during maintenance periods, Orion's processes become starved for CPU time and don't fire off monitoring packets-- literally utilization on the polling interface drops to sometimes as much as 5% of normal monitoring bandwidth. Has anyone experimented with process priority or processor affinity on multi-proc systems?

iunderwo

The problem with the nightly maintenance is that it is processor intensive...enough that a moderately loaded system's polling falls behind.

I have asked if it were possible to move maintenance to another machine, like on the DB server or another external host. That would be best since the summarization does require a fair amount of data crunching and manipulation.

// Ian Underwood - Service Management
// Level 3 Communications

bericks999

I know this response is several years (4 to be exact) old but has anyone else moved the DB to the SAN and did that significantly improve response during high load periods like backups? We have shown that other applications that have been connected to the SAN via 2Gig HBA cards, behave very positively and respond much better when under heavy load then when attached via local SCSI attachments.

Network_Guru

Solarwinds has re-written the DB maintenance routines since these posts, and there are no longer any missed polls during nightly maintenance. I run my DB on a dedicated SQL server with 8Gb RAM (using the AWE switch on x64 OS and hardware) and locally attached Raid 0 disk.
I am upgrading my memory to 14Gb so that the entire DB will run in memory for increased performance.

seudo2k

How did you get the DB ru use more than 4GB? I have a box with 16GB but SQL will only use 4GB. We are running on 2k3 64bit sql 2k5.

Rane

Its been my experience that the nightly job, if running right, does not have an impact on polling. Which is good, because for us it takes 1-2 hours, not the 15 minutes for most of you.

Also, I saw a post farther up about moving your DB to a SAN. While I am not convinced, SW has some good reasons for NOT using a SAN for DB and will chastise you greatly if you do so.

Roger