Single box scalability/sizing?
We currently have ~500 devices/50,000 ports being monitored on a dual cpu E5-2643 v0 system. It's currently running around 80% cpu (all cores w/ hyper-threading) and on a raid-10 16 drive array (sas 2.5" 10K rpm) which averages 1k iops. Most systems are polled in < 5minutes (with about 5-10 taking longer as they're fully loaded boxes).
Looking to plan for future growth, I'm trying to see if anyone is running larger installations and their performance on a single box in trying to gauge if it's worthwhile to spend the $$ to upgrade to a newer V3 cpu system, or if we've effectively reached a single-point saturation limit and have to deploy distributed.
I.e. do we have anyone here running say 1500 devices or 150,000 - 200,000 ports on a single system? If so, care to share your specs?
Thanks.
Steve
Couple of things to scale out from my experience:
- move rrd files to SSD-raid - mount /opt/observium/rrd with noatime - get more cpus, preferably the nice 3.8ghz ones as well - find the right balance your poller threading - move to php7
1500 devices is absolutely possible.
Maarten
Can only agree with Maarten, currently running 1159 devices, 64k ports, 70k sensors on a single box. Specs are: 2x Intel Xeon E5-2630 v3 @ 2.40GHz, 64G RAM and 2x Samsung SM863 SSDs.
/Markus
2016-02-04 10:29 GMT+01:00 Moerman, Maarten mmoerman@ebay.com:
Couple of things to scale out from my experience:
- move rrd files to SSD-raid
- mount /opt/observium/rrd with noatime
- get more cpus, preferably the nice 3.8ghz ones as well
- find the right balance your poller threading
- move to php7
1500 devices is absolutely possible.
Maarten
Maarten Moerman | Mgr, Network Engineering | eBay Classifieds | +31-655122247 | mmoerman@ebay.com
On 2/3/16, 11:40 AM, "observium on behalf of Steve Costaras" < observium-bounces@observium.org on behalf of stevecs@chaven.com> wrote:
We currently have ~500 devices/50,000 ports being monitored on a dual cpu E5-2643 v0 system. It's currently running around 80% cpu (all cores w/ hyper-threading) and on a raid-10 16 drive array (sas 2.5" 10K rpm) which averages 1k iops. Most systems are polled in < 5minutes (with about 5-10 taking longer as they're fully loaded boxes).
Looking to plan for future growth, I'm trying to see if anyone is running larger installations and their performance on a single box in trying to gauge if it's worthwhile to spend the $$ to upgrade to a newer V3 cpu system, or if we've effectively reached a single-point saturation limit and have to deploy distributed.
I.e. do we have anyone here running say 1500 devices or 150,000 - 200,000 ports on a single system? If so, care to share your specs?
Thanks.
Steve
observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium
Thanks, I was looking to see something in the higher range of # of ports. I understand that I can get more write iops with SSD's, or ramdisk. The question really is, on a single dual-cpu system, do we have anyone here that is monitoring in the 150,000 - 200,000 ports range? I.e. If you were to triple your setup, would it still work on a single box?
I note that you are also running two generations newer of cpu's than what we have currently.
On 2016-02-04 03:39, Markus Klock wrote:
Can only agree with Maarten, currently running 1159 devices, 64k ports, 70k sensors on a single box. Specs are: 2x Intel Xeon E5-2630 v3 @ 2.40GHz, 64G RAM and 2x Samsung SM863 SSDs.
/Markus
2016-02-04 10:29 GMT+01:00 Moerman, Maarten <mmoerman@ebay.com mailto:mmoerman@ebay.com>:
Couple of things to scale out from my experience: - move rrd files to SSD-raid - mount /opt/observium/rrd with noatime - get more cpus, preferably the nice 3.8ghz ones as well - find the right balance your poller threading - move to php7 1500 devices is absolutely possible. Maarten -- Maarten Moerman | Mgr, Network Engineering | eBay Classifieds | +31-655122247 <tel:%2B31-655122247> | mmoerman@ebay.com <mailto:mmoerman@ebay.com> On 2/3/16, 11:40 AM, "observium on behalf of Steve Costaras" <observium-bounces@observium.org <mailto:observium-bounces@observium.org> on behalf of stevecs@chaven.com <mailto:stevecs@chaven.com>> wrote: > >We currently have ~500 devices/50,000 ports being monitored on a dual >cpu E5-2643 v0 system. It's currently running around 80% cpu (all >cores w/ hyper-threading) and on a raid-10 16 drive array (sas 2.5" 10K >rpm) which averages 1k iops. Most systems are polled in < 5minutes >(with about 5-10 taking longer as they're fully loaded boxes). > >Looking to plan for future growth, I'm trying to see if anyone is >running larger installations and their performance on a single box in >trying to gauge if it's worthwhile to spend the $$ to upgrade to a newer >V3 cpu system, or if we've effectively reached a single-point saturation >limit and have to deploy distributed. > >I.e. do we have anyone here running say 1500 devices or 150,000 - >200,000 ports on a single system? If so, care to share your specs? > > >Thanks. > >Steve > _______________________________________________ observium mailing list observium@observium.org <mailto: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
participants (3)
-
Markus Klock -
Moerman, Maarten -
Steve Costaras