[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago



I'm still on the 10 way split that I've been on for quite some time, and without my changing anything, our lags got much much better starting on Friday, 4/19 starting with the 12 UTC model sequence. I don't know if this correlated to Unidata switching to a 20 way split or not, but that happened around the same time.

Here are my lag plots, the first ends 04 UTC 4/20, and the second just now at 19 UTC 4/23. Note the Y axis on the first plot goes to ~3600 seconds, but on the second plot, only to ~100 seconds.





Pete



--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: Person, Arthur A. <address@hidden>
Sent: Tuesday, April 23, 2019 1:49 PM
To: Pete Pokrandt; Gilbert Sebenste
Cc: Kevin Goebbert; address@hidden; _NCEP.List.pmb-dataflow; Mike Zuranski; Derek VanPelt - NOAA Affiliate; Dustin Sheffler - NOAA Federal; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 

I switched our test system iddrs2a feeding from conduit.ncep.noaa.gov back to a 2-way split (from a 20-way split) yesterday to see how it would hold up:




While not as good as prior to February, it wasn't terrible, at least until this morning.  Looks like the 20-way split may be the solution going forward if this is the "new normal" for network performance.


                        Art



Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: Pete Pokrandt <address@hidden>
Sent: Saturday, April 20, 2019 12:29 AM
To: Person, Arthur A.; Gilbert Sebenste
Cc: Kevin Goebbert; address@hidden; _NCEP.List.pmb-dataflow; Mike Zuranski; Derek VanPelt - NOAA Affiliate; Dustin Sheffler - NOAA Federal; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
Well, I haven't changed anything in the past few days, but my lags dropped back to pretty much pre-February 10 levels starting with today's (20190419) 12 UTC run. I know Unidata switched to a 20 way split feed around that same time... I am still running a 10-way split. I didn't change anything between today's 06 UTC run and the 12 UTC run, but the lags dropped considerably, and look like they used to.

I wonder if some bad piece of hardware got swapped out somewhere, or if some change was made internally at NCEP that fixed whatever was going on. Or, perhaps the Unidata switch to a 20 way feed somehow reduced a load on a router somewhere and data is getting through more easily?

Strange..

Pete



--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: Person, Arthur A. <address@hidden>
Sent: Thursday, April 18, 2019 2:20 PM
To: Gilbert Sebenste; Pete Pokrandt
Cc: Kevin Goebbert; address@hidden; _NCEP.List.pmb-dataflow; Mike Zuranski; Derek VanPelt - NOAA Affiliate; Dustin Sheffler - NOAA Federal; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 

All --


I switched our test system, iddrs2a, feeding from conduit.ncep.noaa.gov from a 2-way split to a 20-way split yesterday, and the results are dramatic:



Although conduit feed performance at other sites improved a little last night with the MRMS feed failure, it doesn't explain this improvement entirely.  This leads me to ponder the causes of such an improvement:


1) The network path does not appear to be bandwidth constrained, otherwise there would be no improvement no matter how many pipes were used;


2) The problem, therefore, would appear to be packet oriented, either with path packet saturation, or packet shaping.


I'm not a networking expert, so maybe I'm missing another possibility here, but I'm curious whether packet shaping could account for some of the throughput issues.  I've also been having trouble getting timely delivery of our Unidata IDD satellite feed, and discovered that switching that to a 10-way split feed (from a 2-way split) has reduced the latencies from 2000-3000 seconds down to less than 300 seconds.  Interestingly, the peak satellite feed latencies (see below) occur at the same time as the peak conduit latencies, but this path is unrelated to NCEP (as far as I know).  Is it possible that Internet 2 could be packet-shaping their traffic and that this could be part of the cause for the packet latencies we're seeing?


                             Art





Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: address@hidden <address@hidden> on behalf of Gilbert Sebenste <address@hidden>
Sent: Thursday, April 18, 2019 2:29 AM
To: Pete Pokrandt
Cc: Kevin Goebbert; address@hidden; _NCEP.List.pmb-dataflow; Mike Zuranski; Derek VanPelt - NOAA Affiliate; Dustin Sheffler - NOAA Federal; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
FYI: all evening and into the overnight, MRMS data has been missing, QC BR has been town for the last 40 minutes, but smaller products are coming through somewhat more reliably as of 6Z. CONDUIT was still substantially delayed around 4Z with the GFS.

Gilbert

On Apr 16, 2019, at 5:43 PM, Pete Pokrandt <address@hidden> wrote:

Here's a few traceroutes from just now - from idd-agg.aos.wisc.edu to conduit.ncep.noaa.gov. The lags are up and running around 600-800 seconds right now. I'm not including all of the * * * lines from after 144.90.76.65 which is presumably behind a firewall.


2209 UTC Tuesday Apr 16

traceroute -p 388 conduit.ncep.noaa.gov
traceroute to conduit.ncep.noaa.gov (140.90.101.42), 30 hops max, 60 byte packets
 1  vlan-510-cssc-gw.net.wisc.edu (144.92.130.1)  0.906 ms  0.701 ms  0.981 ms
 2  128.104.4.129 (128.104.4.129)  1.700 ms  1.737 ms  1.772 ms
 3  rx-cssc-b380-1-core-bundle-ether2-1521.net.wisc.edu (146.151.168.4)  1.740 ms  3.343 ms  3.336 ms
 4  rx-animal-226-2-core-bundle-ether1-1928.net.wisc.edu (146.151.166.122)  2.043 ms  2.034 ms  1.796 ms
 5  144.92.254.229 (144.92.254.229)  11.530 ms  11.472 ms  11.535 ms
 6  et-1-1-5.4079.rtsw.ashb.net.internet2.edu (162.252.70.60)  22.813 ms  22.899 ms  22.886 ms
 7  et-11-3-0-1275.clpk-core.maxgigapop.net (206.196.177.2)  24.248 ms  24.195 ms  24.172 ms
 8  nwave-clpk-re.demarc.maxgigapop.net (206.196.177.189)  24.244 ms  24.196 ms  24.183 ms
 9  ae-2.666.rtr.clpk.nwave.noaa.gov (137.75.68.4)  24.937 ms  24.884 ms  24.878 ms
10  140.208.63.30 (140.208.63.30)  134.030 ms  126.195 ms  126.305 ms
11  140.90.76.65 (140.90.76.65)  106.810 ms  104.553 ms  104.603 ms

2230 UTC Tuesday Apr 16

traceroute -p 388 conduit.ncep.noaa.gov
traceroute to conduit.ncep.noaa.gov (140.90.101.42), 30 hops max, 60 byte packets
 1  vlan-510-cssc-gw.net.wisc.edu (144.92.130.1)  1.391 ms  1.154 ms  5.902 ms
 2  128.104.4.129 (128.104.4.129)  6.917 ms  6.895 ms  2.004 ms
 3  rx-cssc-b380-1-core-bundle-ether2-1521.net.wisc.edu (146.151.168.4)  3.158 ms  3.293 ms  3.251 ms
 4  rx-animal-226-2-core-bundle-ether1-1928.net.wisc.edu (146.151.166.122)  6.185 ms  2.278 ms  2.425 ms
 5  144.92.254.229 (144.92.254.229)  6.909 ms  13.255 ms  6.863 ms
 6  et-1-1-5.4079.rtsw.ashb.net.internet2.edu (162.252.70.60)  23.328 ms  23.244 ms  28.845 ms
 7  et-11-3-0-1275.clpk-core.maxgigapop.net (206.196.177.2)  30.308 ms  24.575 ms  24.536 ms
 8  nwave-clpk-re.demarc.maxgigapop.net (206.196.177.189)  29.594 ms  24.624 ms  24.618 ms
 9  ae-2.666.rtr.clpk.nwave.noaa.gov (137.75.68.4)  24.581 ms  30.164 ms  24.627 ms
10  140.208.63.30 (140.208.63.30)  25.677 ms  25.767 ms  29.543 ms
11  140.90.76.65 (140.90.76.65)  105.812 ms  105.345 ms  108.857

2232 UTC Tuesday Apr 16

traceroute -p 388 conduit.ncep.noaa.gov
traceroute to conduit.ncep.noaa.gov (140.90.101.42), 30 hops max, 60 byte packets
 1  vlan-510-cssc-gw.net.wisc.edu (144.92.130.1)  1.266 ms  1.070 ms  1.226 ms
 2  128.104.4.129 (128.104.4.129)  1.915 ms  2.652 ms  2.775 ms
 3  rx-cssc-b380-1-core-bundle-ether2-1521.net.wisc.edu (146.151.168.4)  2.353 ms  2.129 ms  2.314 ms
 4  rx-animal-226-2-core-bundle-ether1-1928.net.wisc.edu (146.151.166.122)  2.114 ms  2.111 ms  2.163 ms
 5  144.92.254.229 (144.92.254.229)  6.891 ms  6.838 ms  6.840 ms
 6  et-1-1-5.4079.rtsw.ashb.net.internet2.edu (162.252.70.60)  23.336 ms  23.283 ms  23.364 ms
 7  et-11-3-0-1275.clpk-core.maxgigapop.net (206.196.177.2)  24.493 ms  24.136 ms  24.152 ms
 8  nwave-clpk-re.demarc.maxgigapop.net (206.196.177.189)  24.161 ms  24.173 ms  24.176 ms
 9  ae-2.666.rtr.clpk.nwave.noaa.gov (137.75.68.4)  24.165 ms  24.331 ms  24.201 ms
10  140.208.63.30 (140.208.63.30)  25.361 ms  25.427 ms  25.240 ms
11  140.90.76.65 (140.90.76.65)  113.194 ms  115.553 ms  115.543 ms


2234 UTC Tuesday Apr 16

traceroute -p 388 conduit.ncep.noaa.gov
traceroute to conduit.ncep.noaa.gov (140.90.101.42), 30 hops max, 60 byte packets
 1  vlan-510-cssc-gw.net.wisc.edu (144.92.130.1)  0.901 ms  0.663 ms  0.826 ms
 2  128.104.4.129 (128.104.4.129)  1.645 ms  1.948 ms  1.729 ms
 3  rx-cssc-b380-1-core-bundle-ether2-1521.net.wisc.edu (146.151.168.4)  1.804 ms  1.788 ms  1.849 ms
 4  rx-animal-226-2-core-bundle-ether1-1928.net.wisc.edu (146.151.166.122)  2.011 ms  2.004 ms  1.982 ms
 5  144.92.254.229 (144.92.254.229)  6.241 ms  6.240 ms  6.220 ms
 6  et-1-1-5.4079.rtsw.ashb.net.internet2.edu (162.252.70.60)  23.042 ms  23.072 ms  23.033 ms
 7  et-11-3-0-1275.clpk-core.maxgigapop.net (206.196.177.2)  24.094 ms  24.398 ms  24.370 ms
 8  nwave-clpk-re.demarc.maxgigapop.net (206.196.177.189)  24.166 ms  24.166 ms  24.108 ms
 9  ae-2.666.rtr.clpk.nwave.noaa.gov (137.75.68.4)  24.056 ms  24.306 ms  24.215 ms
10  140.208.63.30 (140.208.63.30)  25.199 ms  25.284 ms  25.351 ms
11  140.90.76.65 (140.90.76.65)  118.314 ms  118.707 ms  118.768 ms

2236 UTC Tuesday Apr 16

traceroute -p 388 conduit.ncep.noaa.gov
traceroute to conduit.ncep.noaa.gov (140.90.101.42), 30 hops max, 60 byte packets
 1  vlan-510-cssc-gw.net.wisc.edu (144.92.130.1)  0.918 ms  0.736 ms  0.864 ms
 2  128.104.4.129 (128.104.4.129)  1.517 ms  1.630 ms  1.734 ms
 3  rx-cssc-b380-1-core-bundle-ether2-1521.net.wisc.edu (146.151.168.4)  1.998 ms  3.437 ms  3.437 ms
 4  rx-animal-226-2-core-bundle-ether1-1928.net.wisc.edu (146.151.166.122)  1.899 ms  1.896 ms  1.867 ms
 5  144.92.254.229 (144.92.254.229)  6.384 ms  6.317 ms  6.314 ms
 6  et-1-1-5.4079.rtsw.ashb.net.internet2.edu (162.252.70.60)  22.980 ms  23.167 ms  23.078 ms
 7  et-11-3-0-1275.clpk-core.maxgigapop.net (206.196.177.2)  24.181 ms  24.152 ms  24.121 ms
 8  nwave-clpk-re.demarc.maxgigapop.net (206.196.177.189)  48.556 ms  47.824 ms  47.799 ms
 9  ae-2.666.rtr.clpk.nwave.noaa.gov (137.75.68.4)  24.166 ms  24.154 ms  24.214 ms
10  140.208.63.30 (140.208.63.30)  25.310 ms  25.268 ms  25.401 ms
11  140.90.76.65 (140.90.76.65)  118.299 ms  123.763 ms  122.207 ms

2242 UTC 

traceroute -p 388 conduit.ncep.noaa.gov
traceroute to conduit.ncep.noaa.gov (140.90.101.42), 30 hops max, 60 byte packets
 1  vlan-510-cssc-gw.net.wisc.edu (144.92.130.1)  1.337 ms  1.106 ms  1.285 ms
 2  128.104.4.129 (128.104.4.129)  6.039 ms  5.778 ms  1.813 ms
 3  rx-cssc-b380-1-core-bundle-ether2-1521.net.wisc.edu (146.151.168.4)  2.275 ms  2.464 ms  2.517 ms
 4  rx-animal-226-2-core-bundle-ether1-1928.net.wisc.edu (146.151.166.122)  2.288 ms  6.978 ms  3.506 ms
 5  144.92.254.229 (144.92.254.229)  10.369 ms  6.626 ms  10.281 ms
 6  et-1-1-5.4079.rtsw.ashb.net.internet2.edu (162.252.70.60)  23.513 ms  23.297 ms  23.295 ms
 7  et-11-3-0-1275.clpk-core.maxgigapop.net (206.196.177.2)  27.938 ms  24.589 ms  28.783 ms
 8  nwave-clpk-re.demarc.maxgigapop.net (206.196.177.189)  28.796 ms  24.630 ms  28.793 ms
 9  ae-2.666.rtr.clpk.nwave.noaa.gov (137.75.68.4)  24.576 ms  24.545 ms  24.587 ms
10  140.208.63.30 (140.208.63.30)  85.763 ms  85.768 ms  83.623 ms
11  140.90.76.65 (140.90.76.65)  131.912 ms  132.662 ms  132.340 ms

Pete

--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: address@hidden <address@hidden> on behalf of Pete Pokrandt <address@hidden>
Sent: Tuesday, April 16, 2019 3:04 PM
To: Gilbert Sebenste; Tyle, Kevin R
Cc: Kevin Goebbert; address@hidden; _NCEP.List.pmb-dataflow; Derek VanPelt - NOAA Affiliate; Mike Zuranski; Dustin Sheffler - NOAA Federal; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
At UW-Madison, we had incomplete 12 UTC GFS data starting with the 177h forecast. Lags exceeded 3600s.



Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: Gilbert Sebenste <address@hidden>
Sent: Tuesday, April 16, 2019 2:44 PM
To: Tyle, Kevin R
Cc: Pete Pokrandt; Dustin Sheffler - NOAA Federal; Mike Zuranski; Derek VanPelt - NOAA Affiliate; Kevin Goebbert; address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
Yes, here at AllisonHouse too...we can feed from a number of sites, and all of them were dropping GFS, and delayed by an hour.

Gilbert

On Apr 16, 2019, at 2:39 PM, Tyle, Kevin R <address@hidden> wrote:

For what it's worth, our 12Z GFS data ingest was quite bad today ... many lost products beyond F168 (we feed from UWisc-MSN primary and PSU secondary).

_____________________________________________
Kevin Tyle, M.S.; Manager of Departmental Computing
Dept. of Atmospheric & Environmental Sciences  
University at Albany
Earth Science 235, 1400 Washington Avenue                       
Albany, NY 12222
Email: address@hidden
Phone: 518-442-4578                            
_____________________________________________

From: address@hidden <address@hidden> on behalf of Pete Pokrandt <address@hidden>
Sent: Tuesday, April 16, 2019 12:00 PM
To: Dustin Sheffler - NOAA Federal; Mike Zuranski
Cc: Kevin Goebbert; address@hidden; Derek VanPelt - NOAA Affiliate; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
All,

Just keeping this in the foreground. 

CONDUIT lags continue to be very large compared to what they were previous to whatever changed back in February. Prior to that, we rarely saw lags more than ~300s. Now they are routinely 1500-2000s at UW-Madison and Penn State, and  over 3000s at Unidata - they appear to be on the edge of losing data. This does not bode well with all of the IDP applications failing back over to CP today..

Can we send you some traceroutes and you back to us to maybe try to isolate where in the network this is happening? It feels like congestion or a bad route somewhere - the lags seem to be worse on weekdays than weekends if that helps at all.

Here are the current CONDUIT lags to UW-Madison, Penn State and Unidata.




<iddstats_CONDUIT_idd_meteo_psu_edu_ending_20190416_1600UTC.gif>






Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: address@hidden <address@hidden> on behalf of Dustin Sheffler - NOAA Federal <address@hidden>
Sent: Tuesday, April 9, 2019 12:52 PM
To: Mike Zuranski
Cc: address@hidden; _NCEP.List.pmb-dataflow; Derek VanPelt - NOAA Affiliate; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
Hi Mike,

Thanks for the feedback on NOMADS. We recently found a slowness issue when NOMADS is running out of our Boulder data center that is being worked on by our teams now that NOMADS is live out of the College Park data center. It's hard sometimes to quantify whether slowness issues that are only being reported by a handful of users is a result of something wrong in our data center, a bad network path between a customer (possibly just from a particular region of the country) and our data center, a local issue on the customers' end, or any other reason that might cause slowness.

Conduit is only ever run from our College Park data center. It's slowness is not tied into the Boulder NOMADS issue, but it does seem to be at least a little bit tied to which of our data centers NOMADS is running out of. When NOMADS is in Boulder along with the majority of our other NCEP applications, the strain on the College Park data center is minimal and Conduit appears to be running better as a result. When NOMADS runs in College Park (as it has since late yesterday) there is more strain on the data center and Conduit appears (based on provided user graphs) to run a bit worse around peak model times as a result. These are just my observations and we are still investigating what may have changed that caused the Conduit latencies to appear in the first place so that we can resolve this potential constraint.

-Dustin

On Tue, Apr 9, 2019 at 4:28 PM Mike Zuranski <address@hidden> wrote:
Hi everyone,

I've avoided jumping into this conversation since I don't deal much with Conduit these days, but Derek just mentioned something that I do have some applicable feedback on...

Two items happened last night.  1. NOMADS was moved back to College Park...

We get nearly all of our model data via NOMADS.  When it switched to Boulder last week we saw a significant drop in download speeds, down to a couple hundred KB/s or slower.  Starting last night, we're back to speeds on the order of MB/s or tens of MB/s.  Switching back to College Park seems to confirm for me something about routing from Boulder was responsible.  But again this was all on NOMADS, not sure if it's related to happenings on Conduit.

When I noticed this last week I sent an email to address@hidden including a traceroute taken at the time, let me know if you'd like me to find that and pass it along here or someplace else.

-Mike

======================
Mike Zuranski
Meteorology Support Analyst
College of DuPage - Nexlab
======================


On Tue, Apr 9, 2019 at 10:51 AM Person, Arthur A. <address@hidden> wrote:

Derek,


Do we know what change might have been made around February 10th when the CONDUIT problems first started happening?  Prior to that time, the CONDUIT feed had been very crisp for a long period of time.


Thanks...            Art



Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: Derek VanPelt - NOAA Affiliate <address@hidden>
Sent: Tuesday, April 9, 2019 11:34 AM
To: Holly Uhlenhake - NOAA Federal
Cc: Carissa Klemmer - NOAA Federal; Person, Arthur A.; Pete Pokrandt; _NCEP.List.pmb-dataflow; address@hidden; address@hidden
Subject: Re: [conduit] Large lags on CONDUIT feed - started a week or so ago
 
Hi all,

Two items happened last night.

1.   NOMADS was moved back to College Park, which means there was a lot more traffic going out which will have effect on the Conduit latencies.  We do not have a full load from the COllege Park Servers as many of the other applications are still running from Boulder, but NOMADS will certainly increase overall load.

2.   As Holly said, there were further issues delaying and changing the timing of the model output yesterday afternoon/evening.  I will be watching from our end, and monitoring the Unidata 48 hour graph (thank you for the link) throughout the day,

Please let us know if you have questions or more information to help us analyse what you are seeing. 

Thank you,

Derek


On Tue, Apr 9, 2019 at 6:50 AM Holly Uhlenhake - NOAA Federal <address@hidden> wrote:
Hi Pete,

We also had an issue on the supercomputer yesterday where several models going to conduit would have been stacked on top of each other instead of coming out in a more spread out fashion.  It's not inconceivable that conduit could have backed up working through the abnormally large glut of grib messages.    Are things better this morning at all?

Thanks,
Holly

On Tue, Apr 9, 2019 at 12:37 AM Pete Pokrandt <address@hidden> wrote:
Something changed starting with today's 18 UTC model cycle, and our lags shot up to over 3600 seconds, where we started losing data. They are growing again now with the 00 UTC cycle as well. PSU and Unidata CONDUIT stats show similar abnormally large lags.

FYI.
Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: Person, Arthur A. <address@hidden>
Sent: Friday, April 5, 2019 2:10 PM
To: Carissa Klemmer - NOAA Federal
Cc: Pete Pokrandt; Derek VanPelt - NOAA Affiliate; Gilbert Sebenste; address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: Large lags on CONDUIT feed - started a week or so ago
 

Carissa,


The Boulder connection is definitely performing very well for CONDUIT.  Although there have been a couple of little blips (~ 120 seconds) since yesterday, overall the performance is superb.  I don't think it's quite as clean as prior to the ~February 10th date when the D.C. connection went bad, but it's still excellent performance.  Here's our graph now with a single connection (no splits):


My next question is:  Will CONDUIT stay pointing at Boulder until D.C. is fixed, or might you be required to switch back to D.C. at some point before that?


Thanks...               Art


Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: Carissa Klemmer - NOAA Federal <address@hidden>
Sent: Thursday, April 4, 2019 6:22 PM
To: Person, Arthur A.
Cc: Pete Pokrandt; Derek VanPelt - NOAA Affiliate; Gilbert Sebenste; address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: Large lags on CONDUIT feed - started a week or so ago
 
Catching up here.

Derek,
Do we have traceroutes from all users? Does anything in VCenter show any system resource constraints?

On Thursday, April 4, 2019, Person, Arthur A. <address@hidden> wrote:

Yeh, definitely looks "blipier" starting around 7Z this morning, but nothing like it was before.  And all last night was clean.  Here's our graph with a 2-way split, a huge improvement over what it was before the switch to Boulder:



Agree with Pete that this morning's data probably isn't a good test since there were other factors.  Since this seems so much better, I'm going to try switching to no split as an experiment and see how it holds up.


                        Art


Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: Pete Pokrandt <address@hidden>
Sent: Thursday, April 4, 2019 1:51 PM
To: Derek VanPelt - NOAA Affiliate
Cc: Person, Arthur A.; Gilbert Sebenste; Anne Myckow - NOAA Affiliate; address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: [Ncep.list.pmb-dataflow] [conduit] Large lags on CONDUIT feed - started a week or so ago
 
Ah, so perhaps not a good test.. I'll set it back to a 5-way split and see how it looks tomorrow.

Thanks for the info,
Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: Derek VanPelt - NOAA Affiliate <address@hidden>
Sent: Thursday, April 4, 2019 12:38 PM
To: Pete Pokrandt
Cc: Person, Arthur A.; Gilbert Sebenste; Anne Myckow - NOAA Affiliate; address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: [Ncep.list.pmb-dataflow] [conduit] Large lags on CONDUIT feed - started a week or so ago
 
HI Pete -- we did have a separate issu hit the CONDUIT feed today.  We should be recovering now, but the backlog was sizeable.  If these numbers are not back to the baseline in the next hour or so please let us know.  We are also watching our queues and they are decreasing, but not as quickly as we had hoped.

Thank you,

Derek

On Thu, Apr 4, 2019 at 1:26 PM 'Pete Pokrandt' via _NCEP list.pmb-dataflow <address@hidden> wrote:
FYI - there is still a much larger lag for the 12 UTC run with a 5-way split compared to a 10-way split. It's better since everything else failed over to Boulder, but I'd venture to guess that's not the root of the problem.



Prior to whatever is going on to cause this, I don'r recall ever seeing lags this large with a 5-way split. It looked much more like the left hand side of this graph, with small increases in lag with each 6 hourly model run cycle, but more like 100 seconds vs the ~900 that I got this morning.

FYI I am going to change back to a 10 way split for now.

Pete



--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: address@hidden <address@hidden> on behalf of Pete Pokrandt <address@hidden>
Sent: Wednesday, April 3, 2019 4:57 PM
To: Person, Arthur A.; Gilbert Sebenste; Anne Myckow - NOAA Affiliate
Cc: address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
Sorry, was out this morning and just had a chance to look into this. I concur with Art and Gilbert that things appear to have gotten better starting with the failover of everything else to Boulder yesterday. I will also reconfigure to go back to a 5-way split (as opposed to the 10-way split that I've been using since this issue began) and keep an eye on tomorrow's 12 UTC model run cycle - if the lags go up, it usually happens worst during that cycle, shortly before 18 UTC each day. 

I'll report back tomorrow how it looks, or you can see at 


Thanks,
Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: address@hidden <address@hidden> on behalf of Person, Arthur A. <address@hidden>
Sent: Wednesday, April 3, 2019 4:04 PM
To: Gilbert Sebenste; Anne Myckow - NOAA Affiliate
Cc: address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 

Anne,


I'll hop back in the loop here... for some reason these replies started going into my junk file (bleh).  Anyway, I agree with Gilbert's assessment.  Things turned real clean around 12Z yesterday, looking at the graphs.  I usually look at flood.atmos.uiuc.edu when there are problem as their connection always seems to be the cleanest.  If there are even small blips or ups and downs in their latencies, that usually means there's a network aberration somewhere that usually amplifies into hundreds or thousands of seconds at our site and elsewhere.  Looking at their graph now, you can see the blipiness up until 12Z yesterday, and then it's flat (except for the one spike around 16Z today which I would ignore):


<pastedImage.png>

Our direct-connected site, which is using a 10-way split right now, also shows a return to calmness in the latencies:


Prior to the recent latency jump, I did not use split requests and the reception had been stellar for quite some time.  It's my suspicion that this is a networking congestion issue somewhere close to the source since it seems to affect all downstream sites.  For that reason, I don't think solving this problem should necessarily involve upgrading your server software, but rather identifying what's jamming up the network near D.C., and testing this by switching to Boulder was an excellent idea.  I will now try switching our system to a two-way split to see if this performance holds up with fewer pipes.  Thanks for your help and I'll let you know what I find out.


                                 Art


Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: address@hidden <address@hidden> on behalf of Gilbert Sebenste <address@hidden>
Sent: Wednesday, April 3, 2019 4:07 PM
To: Anne Myckow - NOAA Affiliate
Cc: address@hidden; _NCEP.List.pmb-dataflow; address@hidden
Subject: Re: [conduit] [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
Hello Anne,

I'll jump in here as well. Consider the CONDUIT delays at UNIDATA:

http://rtstats.unidata.ucar.edu/cgi-bin/rtstats/iddstats_nc?CONDUIT+conduit.unidata.ucar.edu 

And now, Wisconsin: 

http://rtstats.unidata.ucar.edu/cgi-bin/rtstats/iddstats_nc?CONDUIT+idd.aos.wisc.edu 

And finally, the University of Washington:

http://rtstats.unidata.ucar.edu/cgi-bin/rtstats/iddstats_nc?CONDUIT+freshair1.atmos.washington.edu  

All three of whom have direct feeds from you. Flipping over to Boulder definitely caused a major improvement. There was still a brief spike in delay, but much shorter and minimal
compared to what it was.

Gilbert

On Wed, Apr 3, 2019 at 10:03 AM Anne Myckow - NOAA Affiliate <address@hidden> wrote:
Hi Pete,

As of yesterday we failed almost all of our applications to our site in Boulder (meaning away from CONDUIT). Have you noticed an improvement in your speeds since yesterday afternoon? If so this will give us a clue that maybe there's something interfering on our side that isn't specifically CONDUIT, but another app that might be causing congestion. (And if it's the same then that's a clue in the other direction.)

Thanks,
Anne

On Mon, Apr 1, 2019 at 3:24 PM Pete Pokrandt <address@hidden> wrote:
The lag here at UW-Madison was up to 1200 seconds today, and that's with a 10-way split feed. Whatever is causing the issue has definitely not been resolved, and historically is worse during the work week than on the weekends. If that helps at all.

Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: Anne Myckow - NOAA Affiliate <address@hidden>
Sent: Thursday, March 28, 2019 4:28 PM
To: Person, Arthur A.
Cc: Carissa Klemmer - NOAA Federal; Pete Pokrandt; _NCEP.List.pmb-dataflow; address@hidden; address@hidden
Subject: Re: [Ncep.list.pmb-dataflow] Large lags on CONDUIT feed - started a week or so ago
 
Hello Art,

We will not be upgrading to version 6.13 on these systems as they are not robust enough to support the local logging inherent in the new version.

I will check in with my team on if there are any further actions we can take to try and troubleshoot this issue, but I fear we may be at the limit of our ability to make this better.

I’ll let you know tomorrow where we stand. Thanks.
Anne

On Mon, Mar 25, 2019 at 3:00 PM Person, Arthur A. <address@hidden> wrote:

Carissa,


Can you report any status on this inquiry?


Thanks...          Art


Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: Carissa Klemmer - NOAA Federal <address@hidden>
Sent: Tuesday, March 12, 2019 8:30 AM
To: Pete Pokrandt
Cc: Person, Arthur A.; address@hidden; address@hidden; _NCEP.List.pmb-dataflow
Subject: Re: Large lags on CONDUIT feed - started a week or so ago
 
Hi Everyone

I’ve added the Dataflow team email to the thread. I haven’t heard that any changes were made or that any issues were found. But the team can look today and see if we have any signifiers of overall slowness with anything. 

Dataflow, try taking a look at the new Citrix or VM troubleshooting tools if there are any abnormal signatures that may explain this. 

On Monday, March 11, 2019, Pete Pokrandt <address@hidden> wrote:
Art,

I don't know if NCEP ever figured anything out, but I've been able to keep my latencies reasonable (300-600s max, mostly during the 12 UTC model suite) by splitting my CONDUIT request 10 ways, instead of the 5 that I had been doing, or in a single request. Maybe give that a try and see if it helps at all.

Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: Person, Arthur A. <address@hidden>
Sent: Monday, March 11, 2019 3:45 PM
To: Holly Uhlenhake - NOAA Federal; Pete Pokrandt
Cc: address@hidden; address@hidden
Subject: Re: [conduit] Large lags on CONDUIT feed - started a week or so ago
 

Holly,


Was there any resolution to this on the NCEP end?  I'm still seeing terrible delays (1000-4000 seconds) receiving data from conduit.ncep.noaa.gov.  It would be helpful to know if things are resolved at NCEP's end so I know whether to look further down the line.


Thanks...           Art


Arthur A. Person
Assistant Research Professor, System Administrator
Penn State Department of Meteorology and Atmospheric Science
email:  address@hidden, phone:  814-863-1563




From: address@hidden <address@hidden> on behalf of Holly Uhlenhake - NOAA Federal <address@hidden>
Sent: Thursday, February 21, 2019 12:05 PM
To: Pete Pokrandt
Cc: address@hidden; address@hidden
Subject: Re: [conduit] Large lags on CONDUIT feed - started a week or so ago
 
Hi Pete,

We'll take a look and see if we can figure out what might be going on.  We haven't done anything to try and address this yet, but based on your analysis I'm suspicious that it might be tied to a resource constraint on the VM or the blade it resides on.

Thanks,
Holly Uhlenhake
Acting Dataflow Team Lead

On Thu, Feb 21, 2019 at 11:32 AM Pete Pokrandt <address@hidden> wrote:
Just FYI, data is flowing, but the large lags continue.


Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: address@hidden <address@hidden> on behalf of Pete Pokrandt <address@hidden>
Sent: Wednesday, February 20, 2019 12:07 PM
To: Carissa Klemmer - NOAA Federal
Cc: address@hidden; address@hidden
Subject: Re: [conduit] Large lags on CONDUIT feed - started a week or so ago
 
Data is flowing again - picked up somewhere in the GEFS. Maybe CONDUIT server was restarted, or ldm on it? Lags are large (3000s+) but dropping slowly

Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: address@hidden <address@hidden> on behalf of Pete Pokrandt <address@hidden>
Sent: Wednesday, February 20, 2019 11:56 AM
To: Carissa Klemmer - NOAA Federal
Cc: address@hidden; address@hidden
Subject: Re: [conduit] Large lags on CONDUIT feed - started a week or so ago
 
Just a quick follow-up - we started falling far enough behind (3600+ sec) that we are losing data. We got short files starting at 174h into the GFS run, and only got (incomplete) data through 207h.

We have now not received any data on CONDUIT since 11:27 AM CST (1727 UTC) today (Wed Feb 20)

Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden



From: address@hidden <address@hidden> on behalf of Pete Pokrandt <address@hidden>
Sent: Wednesday, February 20, 2019 11:28 AM
To: Carissa Klemmer - NOAA Federal
Cc: address@hidden; address@hidden
Subject: [conduit] Large lags on CONDUIT feed - started a week or so ago
 
Carissa,

We have been feeding CONDUIT using a 5 way split feed direct from conduit.ncep.noaa.gov, and it had been really good for some time, lags 30-60 seconds or less.

However, the past week or so, we've been seeing some very large lags during each 6 hour model suite - Unidata is also seeing these - they are also feeding direct from conduit.ncep.noaa.gov.




Any idea what's going on, or how we can find out? 

Thanks!
Pete


--
Pete Pokrandt - Systems Programmer
UW-Madison Dept of Atmospheric and Oceanic Sciences
608-262-3086  - address@hidden

_______________________________________________
NOTE: All exchanges posted to Unidata maintained email lists are
recorded in the Unidata inquiry tracking system and made publicly
available through the web.  Users who post to any of the lists we
maintain are reminded to remove any personal information that they
do not want to be made public.


conduit mailing list
address@hidden
For list information or to unsubscribe, visit: http://www.unidata.ucar.edu/mailing_lists/


--
Carissa Klemmer
NCEP Central Operations
IDSB Branch Chief
301-683-3835

_______________________________________________
Ncep.list.pmb-dataflow mailing list
address@hidden
https://www.lstsrv.ncep.noaa.gov/mailman/listinfo/ncep.list.pmb-dataflow
--
Anne Myckow
Lead Dataflow Analyst
NOAA/NCEP/NCO
301-683-3825


--
Anne Myckow
Lead Dataflow Analyst
NOAA/NCEP/NCO
301-683-3825
_______________________________________________
NOTE: All exchanges posted to Unidata maintained email lists are
recorded in the Unidata inquiry tracking system and made publicly
available through the web.  Users who post to any of the lists we
maintain are reminded to remove any personal information that they
do not want to be made public.


conduit mailing list
address@hidden
For list information or to unsubscribe, visit: http://www.unidata.ucar.edu/mailing_lists/


--
----
 
Gilbert Sebenste
Consulting Meteorologist
AllisonHouse, LLC
_______________________________________________
Ncep.list.pmb-dataflow mailing list
address@hidden
https://www.lstsrv.ncep.noaa.gov/mailman/listinfo/ncep.list.pmb-dataflow


--
Derek Van Pelt
DataFlow Analyst
NOAA/NCEP/NCO


--
Carissa Klemmer
NCEP Central Operations
IDSB Branch Chief
301-683-3835

_______________________________________________
NOTE: All exchanges posted to Unidata maintained email lists are
recorded in the Unidata inquiry tracking system and made publicly
available through the web.  Users who post to any of the lists we
maintain are reminded to remove any personal information that they
do not want to be made public.


conduit mailing list
address@hidden
For list information or to unsubscribe, visit: http://www.unidata.ucar.edu/mailing_lists/


--
Derek Van Pelt
DataFlow Analyst
NOAA/NCEP/NCO
--
Misspelled straight from Derek's phone.
_______________________________________________
NOTE: All exchanges posted to Unidata maintained email lists are
recorded in the Unidata inquiry tracking system and made publicly
available through the web.  Users who post to any of the lists we
maintain are reminded to remove any personal information that they
do not want to be made public.


conduit mailing list
address@hidden
For list information or to unsubscribe, visit: http://www.unidata.ucar.edu/mailing_lists/
_______________________________________________
Ncep.list.pmb-dataflow mailing list
address@hidden
https://www.lstsrv.ncep.noaa.gov/mailman/listinfo/ncep.list.pmb-dataflow


--
Dustin Sheffler
NCEP Central Operations - Dataflow
5830 University Research Court, Rm 1030
College Park, Maryland 20740
_______________________________________________
NOTE: All exchanges posted to Unidata maintained email lists are
recorded in the Unidata inquiry tracking system and made publicly
available through the web.  Users who post to any of the lists we
maintain are reminded to remove any personal information that they
do not want to be made public.


conduit mailing list
address@hidden
For list information or to unsubscribe, visit: http://www.unidata.ucar.edu/mailing_lists/
_______________________________________________
NOTE: All exchanges posted to Unidata maintained email lists are
recorded in the Unidata inquiry tracking system and made publicly
available through the web.  Users who post to any of the lists we
maintain are reminded to remove any personal information that they
do not want to be made public.


conduit mailing list
address@hidden
For list information or to unsubscribe, visit: 
http://www.unidata.ucar.edu/mailing_lists/