========================================================
E2emonit deployment status
EGEE JRA4 Face to Face meeting, Pisa, 24-25 October 2005
========================================================

Present
-------

UEDIN: Alistair Phipps (AKP), Andy Jackson (AJ)
CNRS:  Sophie Nicoud (SN), Benjamin Vial (BV)
DFN:   Robert Stoy (DFN)

Actions arising
---------------

[BV] Rebuild RPMs from latest source and ask for redeployment at RAL.

[BV] Ensure e2emonit installation guide includes information (or a reference to
information) about tuning the e2emonit tools

[SN] Create script to generate a set of Tier 1 crontabs based on time offset and
measurement interval for each tool

[RS] Determine how to find Path MTU for use in additional pinger tests

Discussion
----------

Deployment status

- There are differences between the versions currently deployed and those tagged
in CVS, e.g. a change in iperf which means measurements are now gathered.  This
requires a redeployment.

Tuning

- Some tools have fixed parameters used in the source code (e.g. the iperf
buffer size).  SA1 may want to change these, and we cannot tune for them as it
depends on the size of the link.  Tuning, or a reference to a document which
describes the tuning, should be included in the install guide.

SA1 Crontabs

- Agreement on initial crontab timings.  These may be changed according to the
needs of SA1 and middleware.

Allow space for 12 tier 1 sites in the crontabs.  SN points out there are only 7
at the moment, but officially there are 11 with 1 more due to come online soon.

These timings are conservative - designed to ensure they interfere minimally
with actual data traffic.  The downside is that with infrequent measurements,
network problems may be harder to track.  SA1 should be able to tune the
crontabs to their own needs.

The timings match those in EDG.

Iperf:
CERN->Site X: iperf every 30 mins, moving to the next Tier 1 site each time.
Assuming 12 Tier 1 sites, this means an iperf test is run to a particular site
every 6 hours.
Site X->CERN: iperf every 6 hours (CERN receives incoming iperf tests every 30
mins).

Udpmon:
This is another bandwidth-intensive test, and will be run at the same frequency
as iperf.

Pinger:
This is less intrusive, so can be run more often.  Note that there are two
pinger tests - one with packet size 100 and one with packet size 1000 - we
assume these are run sequentially and "pinger" in the description relates to
both tests.
CERN->Site X: pinger every 5 mins, moving to the next Tier 1 site each time.
Assuming 12 tier 1 sites, this means an iperf test is run to a particular site
every hour.
Site X->CERN: pinger every hour (CERN receives incoming pinger tests every 5
mins).

   T1 A   T1 B   T1 C
     \     |     /
      \    |    /
       \   |   / \ 30 mins iperf, 5 mins pinger 
T1 H --- Tier 0 --- T1 D
       /   |   \
      /    |    \
     /     |     \
    T1 G  T1 F   T1 E

SN suggests writing a script to generate the crontabs for the Tier 1 sites,
taking the times as a parameter.

RS said it is important to measure pings at path MTU, not just packet sizes of
100 and 1000.  RS will find out how to determine the path MTU for use in tests.

AKP 31/10/2005