Hi Louis,

These images show the r610 as faster, which I'd expect.

I guess that because the R610 is processing things faster, it's putting higher load on the MySQL server, causing the MySQL performance for the UI to tank when the poller is running.

I'd be inclined to run only 8 poller threads on the R610, since it completes in half of the poller period, and it'll keep MySQL performance up for the web UI.

adam.

On 13/08/2015 14:40:48, Louis Bailleul <louis.bailleul@phangos.fr> wrote:
Hi Adam,

I have attached screenshot of both web interfaces with the database metrics popup.

As expected the R610 seems to be a little faster in querying the database.

Best regards,
Louis

August 13 2015 2:36 PM, "Adam Armstrong" wrote:
> For we interface pages, what do the bottom left performance info boxes show?
>
> Screenshots of those might be useful!
>
> Adam.
>
> Sent with AquaMail for Android
> http://www.aqua-mail.com
>
> On 13 August 2015 14:22:48 "Louis Bailleul" wrote:
>
>> Hi Adam,
>>
>> Sorry this is a fairly isolated corporate network.
>> I won't be able to give access from the external world to this boxes.
>>
>> But I am glad to help investigate in anyway I can.
>>
>> Best regards,
>> Louis
>>
>> August 13 2015 12:58 PM, "Adam Armstrong" wrote:
>>> Hi Louis,
>>>
>>> Can you give me http/ssh access to the x3550 to see where the performance drops are?
>>> Thanks,
>>> adam.
>>>
>>>> On 13/08/2015 11:58:42, Louis Bailleul wrote:
>>>>
>>>> Hi,
>>>>
>>>> I am in the process of qualifying some hardware to run Observium (planning to deploy it in
> about
>>> 25
>>>> locations).
>>>>
>>>> So I was trying to evaluate the hardware we currently have which consist of Dell R610 and IBM
>>>> X3550.
>>>> So we can gauge what we need.
>>>>
>>>> The R610 has :
>>>> - 2x Intel X5660 (6 cores 2.8Ghz + hyperthreading) == total of 12 physical cores and 24 logical
>>>> - 96Gb of DDR3 1333Mhz
>>>> - 6x300GB 10K SAS configured as one raid1 of 2 disks for mysql and OS and a raid5 of 4 disks
> for
>>>> rrds.
>>>> - 1Gb network interface
>>>>
>>>> The X3550 has :
>>>> - 2x Intel L5420 (4 cores 2.5Ghz) == total of 8 physical cores
>>>> - 32Gb of DDR2 266Mhz
>>>> - 2x200Gb 7.5K SATA configured as one for mysql and OS and one for rrds
>>>> - 1 Gb network interface
>>>>
>>>> So reading this, it seems easy to guess which one should be faster.
>>>>
>>>> I am currently testing using :
>>>> Observium CE 0.15.6.6430
>>>> OS Linux 2.6.32-358.14.1.el6.x86_64 [amd64] (CentOS 6.2)
>>>> Apache 2.2.15 (CentOS)
>>>> PHP 5.6.11
>>>> Python 2.6.6
>>>> MySQL 5.6.26-log
>>>> SNMP NET-SNMP 5.5
>>>> RRDtool 1.4.9
>>>>
>>>> From my testing, the poller perform a quite better on the R610 than on the X3550 :
>>>> Polling the same 310 linux boxes (The slowest devices taking around 35 seconds) this gives an
>>>> average of :
>>>> - 8 threads :
>>>> R610 : 169 seconds
>>>> X3550 : 213 seconds
>>>> - 12 threads :
>>>> R610 : 69 seconds
>>>> X3550 : 156 seconds
>>>> - 24 threads
>>>> R610 : 41 seconds
>>>> X3550 : 107 seconds
>>>> - 36 threads
>>>> R610 : 41 seconds
>>>> X3550 : 160 seconds
>>>> - 64 threads
>>>> R610 : 37 seconds
>>>> X3550 : 225 seconds
>>>> - 96 Threads
>>>> R610 : 40 seconds
>>>> X3550 : Still running
>>>>
>>>> Ok we clearly have a winner here.
>>>> Bottom line for the poller, you can come close to the time taken by your slowest devices if you
>>> add
>>>> more threads.
>>>> But it seems that you need quite a few cores available to be able to increase the number of
>>> threads
>>>> without decreasing the efficiency.
>>>>
>>>> This was kind of expected.
>>>>
>>>> Where the thing becomes weird it's on the web interface and especially the graph rendering :
>>>> On average starting with a clear cache, the same Linux device page render (graphs included)
>>>> consistently in about 1 second on the X3550 and in about 4 seconds on the R610 (screenshot
>>>> attached).
>>>>
>>>> Where it became even worse : if you ask the X3550 to connect to the mysql database of the R610
>>> and
>>>> export the rrds of the R610 over NFS ... the X3550 still render the page faster (still around 1
>>>> second).
>>>>
>>>> Any guess on what is going on here ?
>>>> It doesn't seem right that a machine that is definitely faster accessing mysql and updating
> rrds
>>>> would be 4 times slower when rendering graphs.
>>>>
>>>> Best regards,
>>>> Louis
>>>> _______________________________________________
>>>> 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
_______________________________________________
observium mailing list
observium@observium.org
http://postman.memetic.org/cgi-bin/mailman/listinfo/observium