Hi,
Sorry for the delay but I couldn't test it before.
I finally managed to replace rrdtool 1.4.7 by rrdtool 1.4.9.
And I am glad I did!
There is indeed something broken with 1.4.7 because the speed up is quite amazing.
The web interface is now blazing fast, even for my switches with 700+ ports. Pages which used to render in between 8 and 12min now are displayed in 34 seconds.
Thanks a lot for pointing this out and keep up the good work on this amazing tool.
Best regards, Louis
January 13 2015 6:59 PM, "Adam Armstrong" adama@memetic.org wrote:
If rrd write performance (poller runtime) is an issue, you might benefit from rrdcached (or SSD/ramdisk storage).
If rrd read performance is an issue, you will be better served by ssd or ramdisk storage. rrdcached /might/ give some performance improvements from reducing write i/o load, but the problem with rrdcached is that in order to read an rrd you need to signal to rrdcached to flush its cache so you can read the rrd, adding an extra step, which may not be trivial in terms of performance on our graph-heavy pages.
I still say that the best performance comes from ramdisk, with ssd an option too.
I've run all of my Observium systems from ramdisk for years, with automated backup/restore scripts. If I reboot a server, I only lose data since the last backup (backing up rrd data with fast compression is pretty quick, so you can run it fairly frequently).
My dislike of rrdcached comes from it never really helping in any of the instances i had performance problems. Sure, it makes polling slightly quicker, but it doesn't really do anything for the loading of graphs, which is where the human<>computer irritation comes from.
adam.
On 2015-01-13 00:36, Tristan Rhodes wrote:
This is all good info. I haven't used rrdcached before but I think I will give it a shot.
Regarding writes versus reads, aren't there many, many times the amount of writes compared to reads?
For instance, the RRDs for all components on all devices are written every 5 mins, correct?
Reads only occur when someone is browsing the website.
Am I correct about this?
Cheers,
Tristan
TRISTAN RHODES Network Engineer Weber State University 801.626.8549
On Mon, Jan 12, 2015 at 3:41 PM, Adam Armstrong adama@memetic.org wrote:
Not that rrdcached only speeds up RRD writing, which is why we don't really use it ourselves.
It actually slows down reading (generating graphs), too!
adam.
On 2015-01-12 10:42, Markus Klock wrote:
The ramdisk itself is not very reliable no, the trick is to make a backup of the ramdisk to a reglar harddrive or SSD once in a while so that in a case of power outage or unexpected reboot you could just copy the data back to the ramdisk from the harddrive.
An option to ramdisk would be to use rrdcached. In large enviroments rrdcached could cut I/O in half or more. A quick guide can be found here: http://blog.best-practice.se/2014/10/using-rrdcached-with-observium.html [1] [2]
SSD disks + rrdcached could handle a pretty big enviroment. But using a ramdisk is still very very much faster.
/Markus
2015-01-12 10:52 GMT+01:00 Louis Bailleul louis.bailleul@phangos.fr:
Hi Adam ,
Is ramdisk storage reliable enough ? Not that I have no faith in the reliability of our setup, but storing large amount of data in ram isn't a little risky ?
Currently our rrd dir is du -sh rrd/ 21G rrd/
That fit in a 24-32Go ram VM which is something that we can create.
What do you think of the mysql tuning that have been suggested by Eduardo and Alastair ?
Is there any other option than ramdisk storage ?
Best regards, Louis
January 9 2015 5:01 PM, "Adam Armstrong" adama@memetic.org wrote:
Depending upon your deployment size, I usually recommend ramdisk storage for RRDs.
adam.
------ Original Message ------ From: "Louis Bailleul" louis.bailleul@phangos.fr To: "Observium Network Observation System" observium@observium.org Sent: 1/9/2015 10:51:40 AM Subject: Re: [Observium] Speed up graph rendering
Hi,
Thank you for the pointers.
To be more specific the page generation for an arista switch with 760 ports take 0.567 seconds. But the loading of all graphs (a lot of sensors) take 8m 53s according to firebug.
Generating the ports view page takes 1.421s but loading all the graphs takes 12m 33s and 493 HTTP request (maybe not accurate as firebug says the log limit was reached).
On a force10 switch with 48 ports page generates in 0.207seconds and the page loads in 2.73s. The ports view takes 0.311 seconds to generate and the page finishes loading graphs after 1m8s (205 HTTP requests in total).
So from what I understand the mysql speed might not be in cause but maybe the RRD disk I/O. What can I do to speed it up ? Moving the disk on faster storage will be enough ?
Could it be that apache need tuning to handle that amount of call to graph.php ?
For the record the disk system is an IBM V3700 SAS attached to the ESXi hosts.
Best regards, Louis
January 9 2015 3:11 PM, "Tom Laermans" mailto:%22Tom%20Laermans%22%20%3Ctom.laermans@powersource.cx%3E wrote:
Hi,
You must note however, that the load time for the page as displayed by Observium is the generation time of the HTML (which is influenced by MySQL speed and CPU, mostly), which includes image tags to load the graphs - but not the loading of the graphs themselves (which is heavily influenced by both CPU and RRD Disk I/O).
Tom
On 01/09/2015 04:08 PM, Pedersen, Sean wrote: It might help if you included a couple of references/examples of the pages that are slow.
To give you something to compare against, we’re running Observium on an ESX VM – 4x Xeon X5690 @ 3.5GHz, 8GB RAM, no idea on storage – but we’re at a pretty consistent 40-50% utilization at all times, using about 2GB of RAM. This is with ~200 devices and 16K ports.
I looked at a Cisco switch stack that has 115 active ports. Graphs > System loaded in .45 seconds, which is only 16 different graphs between four categories. Likewise, the Ports page loaded at .64 seconds with the miniature preview graphs. Clicking on a specific port, it loaded 24 graphs in .49 seconds.
Now…I have some custom port lists, one of which is 36 graphs, which loaded in .53 seconds. However, I have another group that has…a metric shit-tonne of ports and graphs on it – hundreds of them. The page loaded in 1.07 seconds, but it took several seconds more for all of the individual graphs to populate, which I expect considering the sheer number of them present in that page. I don’t know if that’s simply the browser catching up, or if it’s still generating the actual graphs in the background.
From: Louis Bailleul louis.bailleul@phangos.fr Reply-To: "observium@observium.org" observium@observium.org Date: Friday, January 9, 2015 at 7:46 AM To: "observium@observium.org" observium@observium.org Subject: [Observium] Speed up graph rendering
Hi,
I am currently using Observium on 149 switches with a total of 10173 ports. I can't say that anything is wrong with it, but the web interface seems a bit slow especially when you want to display pages with a lot of graphs.
Observium is currently running in a VM with 4 xeon E5-2650 cores and 3Gb of ram. It's virtual disks are on a local raid5 composed of 10K SAS disks.
Observium report that it's VM use on average 25% of its CPUs with spikes at 40% when the web interface is heavily used. The disk I/O doesn't go past 150 ops OUT and rare spikes at 50 IN.
Anyone could advise on what kind of hardware or software tweaking could help speed up the thing ?
Are they special requirements/tunnings for the rrd storage or the mysql database ?
Best regards, Louis
FOUNDED IN 2007, IO PROVIDES THE DATA CENTER AS A SERVICE TO BUSINESSES AND GOVERNMENTS AROUND THE WORLD.
The communication contained in this e-mail is confidential and is intended only for the named recipient(s) and may contain information that is privileged, proprietary, attorney work product or exempt from disclosure under applicable law. If you have received this message in error, or are not the named recipient(s), please note that any form of distribution, copying or use of this communication or the information in it is strictly prohibited and may be unlawful. Please immediately notify the sender of the error, and delete this communication including any attached files from your system. Thank you for your cooperation.
observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium [2] [1]
observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium [2] [1]
Links:
[1] http://postman.memetic.org/cgi-bin/mailman/listinfo/observium [2] [2] http://blog.best-practice.se/2014/10/using-rrdcached-with-observium.html [1]
observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium [2]
observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium [2]
Links:
[1] http://blog.best-practice.se/2014/10/using-rrdcached-with-observium.html [2] http://postman.memetic.org/cgi-bin/mailman/listinfo/observium
observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium
observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium