Frontier T0 Squids overload of August 7th, 2014
Contents
Introduction
A set of workflows related to Cosmics data was set to run in the Agile Infrastructure at CERN (
T2_CH_CERN_AI
), which utilices the T0 squids
cmst0frontier{1,2}
as specified in its
site-local-config.xml
. An overview of the spill-over activity from the squids onto the central Launchpads can be seen below:
Given the completeness of a squid log and the sheer amount of queries a Frontier squid usually undergoes, the statistics to be reported in this document focus on specific time spans, over which the logs are aggregated in order to extract the most relevant patterns. The chosen time spans (CET time zone) at the time of this writing were:
- Just before Noon: From 11:00 AM to 12:00 PM
- Early Evening: From 6:00 PM to 7:00 PM
Situation under heavy load: Just before Noon
Since on this day the
site-local-config.xml
included the clause for the backup proxies
cmsbpfrontier{1,2}
, the traffic statistics for the T0 squids and the backup proxies are compared.
The biggest shares of the load (measured by data transferred) per Frontier ID (which traces the kind of job that made the Frontier query) is shown below:
Query type |
Frontier ID |
Share [%] |
PromptProd |
wmagent_PromptReco_Run224187_MinimumBias |
87.94 |
wmagent_PromptReco_Run224409_Cosmics |
3.52 |
wmagent_PromptReco_Run224413_MinimumBias |
1.69 |
wmagent_PromptReco_Run224259_MinimumBias |
1.62 |
wmagent_PromptReco_Run224471_Cosmics |
1.56 |
Others |
3.68 |
FrontierProd |
CMSSW_7_0_1 |
40.98 |
wmagent_jbadillo_ACDC_BTV-Fall13dr-00210_T1_US_FNAL_MSS_00084_v0_castor_tsg_140806_141814_7313 |
36.20 |
wmagent_alahiff_BTV-Spring14dr-00120_T1_US_FNAL_MSS_00120_v0_castor_140806_132554_7742 |
4.35 |
CMSSW_7_2_X_2014-08-07-0200 |
3.52 |
CMSSW_7_2_DEVEL_X_2014-08-07-0200 |
2.37 |
Others |
12.58 |
Regarding the origin regions of the queries, here is the distribution of them (again, measured by data transferred)
IP range |
PromptProd Share [%] |
FrontierProd Share [%] |
128.*.*.* |
77.92 |
91.25 |
188.*.*.* |
21.79 |
7.97 |
137.*.*.* |
0.28 |
0.78 |
127.0.0.1 |
0.01 |
0.00 |
It is interesting to note that for the
PromptProd
queries, a significant share of them were issued from Wigner (IP 188.*), whereas most of the
FrontierProd
queries where from Meyrin (the other IP ranges)
Situation under normal load: Early Evening
In writing
--
LuisLinares - 08 Aug 2014