All, Isn’t 95th @ 1m unusual and non-standard? http://micelg.usinternet.com/export/graph_385.html Best, -M<
a secant line is a secant line, be it a ray/segment/line/etc between points spaced 1 second, ten seconds, or ten minutes apart, no? (sorting line segments by their n-tile magnitude, be it at 95% or some other interval, seems decoupled from the measurement/secant interval) if my graph theory is bad, let me know :) -Tk On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com> wrote:
All,
Isn’t 95th @ 1m unusual and non-standard?
http://micelg.usinternet.com/export/graph_385.html
Best,
-M<
Doesn't more sampling make it appear as if there is more traffic? The de facto std for IXPs is 5m polling.
On Aug 12, 2015, at 12:48, Anton Kapela <tkapela@GMAIL.COM> wrote:
a secant line is a secant line, be it a ray/segment/line/etc between points spaced 1 second, ten seconds, or ten minutes apart, no?
(sorting line segments by their n-tile magnitude, be it at 95% or some other interval, seems decoupled from the measurement/secant interval)
if my graph theory is bad, let me know :)
-Tk
On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com> wrote:
All,
Isn’t 95th @ 1m unusual and non-standard?
http://micelg.usinternet.com/export/graph_385.html
Best,
-M<
Depends on how you look at it. More sampling shows more extremes. If you sample at 1 minute, and it were to shoot up to 12G for a minute or two, and then drop below 1G for a minute or two, you would see both the 12G high, and the 1G low. If you look at the same traffic on a five min sample rate, you would see it at about 6G, and you would not see the 1G low, or the 12G high. Personally I like higher sample rates since if you are sizing a connection to the exchange, and you do not want to have any loss, in the above example you would have loss if you only purchased a 10G circuit, and if you were looking only at 5 minute sample rates, you would have no clue why you were getting intermittent packet loss. -----Original Message----- From: MICE Discuss [mailto:MICE-DISCUSS@LISTS.IPHOUSE.NET] On Behalf Of Hannigan, Martin Sent: Wednesday, August 12, 2015 12:24 PM To: MICE-DISCUSS@LISTS.IPHOUSE.NET Subject: Re: [MICE-DISCUSS] SNMP Measurements for MICE? Doesn't more sampling make it appear as if there is more traffic? The de facto std for IXPs is 5m polling.
On Aug 12, 2015, at 12:48, Anton Kapela <tkapela@GMAIL.COM> wrote:
a secant line is a secant line, be it a ray/segment/line/etc between points spaced 1 second, ten seconds, or ten minutes apart, no?
(sorting line segments by their n-tile magnitude, be it at 95% or some other interval, seems decoupled from the measurement/secant interval)
if my graph theory is bad, let me know :)
-Tk
On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com> wrote:
All,
Isn’t 95th @ 1m unusual and non-standard?
http://micelg.usinternet.com/export/graph_385.html
Best,
-M<
Sure, but why does that matter with respect to the IXP fabric? It’s misleading to some extent, along the lines you’ve explained. The idea behind IXP statistics is about volume. Networks typically use it as a gauge of value and especially when there are two IXPs competing in a market. The 5m sampling benchmark is also not a de-facto standard, it’s a fully fledged standard. It is nice that the graph is clear about the sample rate though so not a complaint, pointing out an oddity as contrasted to 99% of the IXP world. There’s really no need for the MICE to be doing 1m IMHO. Best, -M<
On Aug 12, 2015, at 1:47 PM, Jeremy Lumby <jlumby@MNVOIP.COM> wrote:
Depends on how you look at it. More sampling shows more extremes. If you sample at 1 minute, and it were to shoot up to 12G for a minute or two, and then drop below 1G for a minute or two, you would see both the 12G high, and the 1G low. If you look at the same traffic on a five min sample rate, you would see it at about 6G, and you would not see the 1G low, or the 12G high. Personally I like higher sample rates since if you are sizing a connection to the exchange, and you do not want to have any loss, in the above example you would have loss if you only purchased a 10G circuit, and if you were looking only at 5 minute sample rates, you would have no clue why you were getting intermittent packet loss.
-----Original Message----- From: MICE Discuss [mailto:MICE-DISCUSS@LISTS.IPHOUSE.NET] On Behalf Of Hannigan, Martin Sent: Wednesday, August 12, 2015 12:24 PM To: MICE-DISCUSS@LISTS.IPHOUSE.NET Subject: Re: [MICE-DISCUSS] SNMP Measurements for MICE?
Doesn't more sampling make it appear as if there is more traffic? The de facto std for IXPs is 5m polling.
On Aug 12, 2015, at 12:48, Anton Kapela <tkapela@GMAIL.COM> wrote:
a secant line is a secant line, be it a ray/segment/line/etc between points spaced 1 second, ten seconds, or ten minutes apart, no?
(sorting line segments by their n-tile magnitude, be it at 95% or some other interval, seems decoupled from the measurement/secant interval)
if my graph theory is bad, let me know :)
-Tk
On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com> wrote:
All,
Isn’t 95th @ 1m unusual and non-standard?
http://micelg.usinternet.com/export/graph_385.html
Best,
-M<
On Wed, Aug 12, 2015 at 12:47:00PM -0500, Jeremy Lumby wrote:
Depends on how you look at it. More sampling shows more extremes.
Won't matter. Unless someone changed things in the RRD files they will all get rolled up anyway and the spikes and dips will then just get rounded out. Want to not have that happen? Use RTG2. Can also poll 1700+ interfaces in under a minute, probably faster on todays hardware. -- Mike Horwath, reachable via drechsau@Geeks.ORG
If we look closely at the 1 minute graph-- the steps are actually in 5 minute chunks. (It's even more apparent if you log into the Cacti site and zoom in). I'm pretty sure it's only polling every 5 minutes, but the 1 minute Cacti template is enabled and being exported. A person could probably pretend the top 1 minute graph doesn't exist, since the 1 minute data appears to be interpolated. I'm not sure I understand the 95th (percentile) reference, though? ~Matthew *Matthew Beckwell* - V.P. - Network Operations Advanced Integrated Technologies Office: +1.952.829.5511 x204 E-Mail: matthewb@aitech.net On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com> wrote:
All,
Isn’t 95th @ 1m unusual and non-standard?
http://micelg.usinternet.com/export/graph_385.html
Best,
-M<
My point was about the polling interval. Thanks! -M< On Aug 12, 2015, at 2:05 PM, Matthew Beckwell <matthewb@AITECH.NET<mailto:matthewb@AITECH.NET>> wrote: If we look closely at the 1 minute graph-- the steps are actually in 5 minute chunks. (It's even more apparent if you log into the Cacti site and zoom in). I'm pretty sure it's only polling every 5 minutes, but the 1 minute Cacti template is enabled and being exported. A person could probably pretend the top 1 minute graph doesn't exist, since the 1 minute data appears to be interpolated. I'm not sure I understand the 95th (percentile) reference, though? ~Matthew Matthew Beckwell - V.P. - Network Operations Advanced Integrated Technologies Office: +1.952.829.5511 x204 E-Mail: matthewb@aitech.net<mailto:matthewb@aitech.net> On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com<mailto:marty@akamai.com>> wrote: All, Isn’t 95th @ 1m unusual and non-standard? http://micelg.usinternet.com/export/graph_385.html<https://urldefense.proofpoint.com/v2/url?u=http-3A__micelg.usinternet.com_export_graph-5F385.html&d=BQMFaQ&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=5QAtDTZgYiOXg9A8YPMItc1F8NuMPAX_dfDso12OBtc&s=jMC_yIaEGBaJpqz22z_jHgZTgOThSd2lI3BeGF2vGJo&e=> Best, -M< ________________________________ To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1<https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.iphouse.net_cgi-2Dbin_wa-3FSUBED1-3DMICE-2DDISCUSS-26A-3D1&d=BQMFaQ&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=5QAtDTZgYiOXg9A8YPMItc1F8NuMPAX_dfDso12OBtc&s=J9BgB3E03TDVao9_EbXKzMN75CGEx5Cws4bobY04C5s&e=>
The cacti server is polling on 5 minute interval. ________________________________ From: MICE Discuss [MICE-DISCUSS@LISTS.IPHOUSE.NET] on behalf of Hannigan, Martin [marty@AKAMAI.COM] Sent: Wednesday, August 12, 2015 1:10 PM To: MICE-DISCUSS@LISTS.IPHOUSE.NET Subject: Re: [MICE-DISCUSS] SNMP Measurements for MICE? My point was about the polling interval. Thanks! -M< On Aug 12, 2015, at 2:05 PM, Matthew Beckwell <matthewb@AITECH.NET<mailto:matthewb@AITECH.NET>> wrote: If we look closely at the 1 minute graph-- the steps are actually in 5 minute chunks. (It's even more apparent if you log into the Cacti site and zoom in). I'm pretty sure it's only polling every 5 minutes, but the 1 minute Cacti template is enabled and being exported. A person could probably pretend the top 1 minute graph doesn't exist, since the 1 minute data appears to be interpolated. I'm not sure I understand the 95th (percentile) reference, though? ~Matthew Matthew Beckwell - V.P. - Network Operations Advanced Integrated Technologies Office: +1.952.829.5511 x204 E-Mail: matthewb@aitech.net<mailto:matthewb@aitech.net> On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com<mailto:marty@akamai.com>> wrote: All, Isn’t 95th @ 1m unusual and non-standard? http://micelg.usinternet.com/export/graph_385.html<https://urldefense.proofpoint.com/v2/url?u=http-3A__micelg.usinternet.com_export_graph-5F385.html&d=BQMFaQ&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=5QAtDTZgYiOXg9A8YPMItc1F8NuMPAX_dfDso12OBtc&s=jMC_yIaEGBaJpqz22z_jHgZTgOThSd2lI3BeGF2vGJo&e=> Best, -M< ________________________________ To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1<https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.iphouse.net_cgi-2Dbin_wa-3FSUBED1-3DMICE-2DDISCUSS-26A-3D1&d=BQMFaQ&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=5QAtDTZgYiOXg9A8YPMItc1F8NuMPAX_dfDso12OBtc&s=J9BgB3E03TDVao9_EbXKzMN75CGEx5Cws4bobY04C5s&e=> ________________________________ To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1
You can see here that the reporting claims it’s 1M http://micelg.usinternet.com/export/graph_385.html Best, -M< On Aug 12, 2015, at 2:36 PM, Justin Krejci <JKrejci@USINTERNET.COM<mailto:JKrejci@USINTERNET.COM>> wrote: The cacti server is polling on 5 minute interval. ________________________________ From: MICE Discuss [MICE-DISCUSS@LISTS.IPHOUSE.NET<mailto:MICE-DISCUSS@LISTS.IPHOUSE.NET>] on behalf of Hannigan, Martin [marty@AKAMAI.COM<mailto:marty@AKAMAI.COM>] Sent: Wednesday, August 12, 2015 1:10 PM To: MICE-DISCUSS@LISTS.IPHOUSE.NET<mailto:MICE-DISCUSS@LISTS.IPHOUSE.NET> Subject: Re: [MICE-DISCUSS] SNMP Measurements for MICE? My point was about the polling interval. Thanks! -M< On Aug 12, 2015, at 2:05 PM, Matthew Beckwell <matthewb@AITECH.NET<mailto:matthewb@AITECH.NET>> wrote: If we look closely at the 1 minute graph-- the steps are actually in 5 minute chunks. (It's even more apparent if you log into the Cacti site and zoom in). I'm pretty sure it's only polling every 5 minutes, but the 1 minute Cacti template is enabled and being exported. A person could probably pretend the top 1 minute graph doesn't exist, since the 1 minute data appears to be interpolated. I'm not sure I understand the 95th (percentile) reference, though? ~Matthew Matthew Beckwell - V.P. - Network Operations Advanced Integrated Technologies Office: +1.952.829.5511 x204 E-Mail: matthewb@aitech.net<mailto:matthewb@aitech.net> On Wed, Aug 12, 2015 at 9:11 AM, Hannigan, Martin <marty@akamai.com<mailto:marty@akamai.com>> wrote: All, Isn’t 95th @ 1m unusual and non-standard? http://micelg.usinternet.com/export/graph_385.html<https://urldefense.proofpoint.com/v2/url?u=http-3A__micelg.usinternet.com_export_graph-5F385.html&d=BQMFaQ&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=5QAtDTZgYiOXg9A8YPMItc1F8NuMPAX_dfDso12OBtc&s=jMC_yIaEGBaJpqz22z_jHgZTgOThSd2lI3BeGF2vGJo&e=> Best, -M< ________________________________ To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1<https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.iphouse.net_cgi-2Dbin_wa-3FSUBED1-3DMICE-2DDISCUSS-26A-3D1&d=BQMFaQ&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=5QAtDTZgYiOXg9A8YPMItc1F8NuMPAX_dfDso12OBtc&s=J9BgB3E03TDVao9_EbXKzMN75CGEx5Cws4bobY04C5s&e=> ________________________________ To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1<https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.iphouse.net_cgi-2Dbin_wa-3FSUBED1-3DMICE-2DDISCUSS-26A-3D1&d=BQMF-g&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=Um3915PmjHlwqMcq98O_i6mPerdOv0Mjn1f8gx4xPl4&s=_i-23SZIWRefuI-FdBoV8l_g08sob08wtVZ6gSaB_2c&e=> ________________________________ To unsubscribe from the MICE-DISCUSS list, click the following link: http://lists.iphouse.net/cgi-bin/wa?SUBED1=MICE-DISCUSS&A=1<https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.iphouse.net_cgi-2Dbin_wa-3FSUBED1-3DMICE-2DDISCUSS-26A-3D1&d=BQMF-g&c=96ZbZZcaMF4w0F4jpN6LZg&r=XDN_BIPGnpb6V0w5M9FADw&m=Um3915PmjHlwqMcq98O_i6mPerdOv0Mjn1f8gx4xPl4&s=_i-23SZIWRefuI-FdBoV8l_g08sob08wtVZ6gSaB_2c&e=>
On 08/12/2015 01:39 PM, Hannigan, Martin wrote:
You can see here that the reporting claims it’s 1M
Even if it was polling at 1 minute (which from other messages in this thread, we seem to think it isn't), that's just for the hourly graph. The daily graph is unambiguously claiming 5 minute averages. In light of that, is this still a problem? -- Richard
On Aug 12, 2015, at 2:42 PM, Richard Laager <rlaager@WIKTEL.COM> wrote:
On 08/12/2015 01:39 PM, Hannigan, Martin wrote:
You can see here that the reporting claims it’s 1M
Even if it was polling at 1 minute (which from other messages in this thread, we seem to think it isn't), that's just for the hourly graph. The daily graph is unambiguously claiming 5 minute averages.
In light of that, is this still a problem?
It wasn’t really a problem more than a curiosity. Even if it was polling at 1m, as long as the statistics collection method is noted clearly and publicly it’s shouldn’t be a big deal. Best, -M<
In light of that, is this still a problem?
It wasn't really a problem more than a curiosity. Even if it was polling at 1m, as long as the statistics collection method is noted clearly and publicly it's shouldn't be a big deal.
The hourly graph is just some default view that got turned on and is being exported with everything else, if the group thinks it would somehow be more clear to remove it we could do that - as several have noted its increments are (correctly) 5 minutes. In other graph news, I think we've got the Peers/Prefixs graphs displaying correctly now and I think they look pretty good - thanks for people's help troubleshooting that. They're under the "Route Server" portion of the graph tree. Cheers, anthony
participants (8)
-
Anthony Anderberg -
Anton Kapela -
Hannigan, Martin -
Jeremy Lumby -
Justin Krejci -
Matthew Beckwell -
Mike Horwath -
Richard Laager