Level 3 Online Status Reports

Issue 13
Monday, July 19, 1999

Event Flow

Periodic push ATM driver: The 80ms delay has been traced to the printk() routine built with serial console echoing: an entire line was being sent via the serial port at 9600bps. The serial console echoing has been removed for now, though it may be useful later on. It now appears that the driver operates satisfactorily, even with data being sent from the processor nodes to the output node without event rejection. Real-time mode was not used for these tests.
Event rates with new driver, with 4 processor nodes writing to trash, 10 SCPU's sending 12KB each: 87.7 ev/s, the same as previous similar tests. No indications of the driver, with its 5ms poll, slowing things down.
Test wedge has been fixed but has been busy.
From the PCFARMS meeting: the Nightshade (N440BX) motherboard is being discontinued at the end of this quarter. Intel recommends the "Lancewood" (L440GX), one of which was received by Don last Thursday. It includes better support for Pentium III. Don has tested the IPMI (on-board sensor readout) code, which works, as do 2.0 and 2.2 kernels. However, the board he received is a single- processor board, so there were no tests of SMP. Also missing are netperf tests.

Event builder monitoring code: the memory leak has been fixed. Overnight tests of the SM control (~12 hours, ~2M calls) were stopped manually and showed no problems.
The scanner manager control/monitoring code has been compiled with Boris's code for reading the scanner manager IOR out of the database. This works.
Boris is reported to have a flat-file implementation. This can also be used to load the default values prior to any other database (or other flat-file) access.
Toy Run Control proxy for event builder/Level 3: Alexander will implement this with initiation and help from Steve.
Resource database: Geri presented a diagram at last Friday's DAQ meeting representing the relationships within the resource manager's database. She has now added relationships with SCPU's and Level 3 nodes.

xntp has been implemented across the Level 3 farm, with the exception of b0l3002 and b0l3003 (old nodes), which have older versions.
Low-level monitoring primitives:
- Andreas has implemented more /proc accesses and can also now access network statistics.
- Ivan has hacked Don's IPMI code to read on-board sensors for CPU temperature, fan speed, and voltages.
Relaying code: Ivan can now execute scripts from his relay code, thus opening the way to executing Level 3 processes.
Histograms: Mike has demonstrated the ability to book histograms and circular buffers, fill them, and unpack them with ROOT. The input data is faked. He now needs to integrate it with Ivan's relay mechanism.
Integration: Ivan and Mike will attempt to monitor CPU temperature using their packages to fill, transmit, and display the data. The first attempt entailed updating Mike's code to compile with egcs and link with ROOT v2.21/01 (the last version built with egcs 1.0.2, while we have egcs 1.0.3a installed). The histogramming package and the relay software has been made to work, but the sensor readout has not.
Andreas is attending the SmartSockets class. A question was raised as to whether Andreas should pass on some of the SmartSockets wisdom to Alexander so he can work on it while Andreas is away studying.
We have to keep in mind that we will be creating a GUI for steady-state monitoring.

Peter Wilson says that the power supplies for the 1st floor racks have just started to arrive.
We will wait for the 20 duplex fibers to arrive, see if they will work, and then order the other optical fibers for the ATM-converter connections and SCRAMNET, which are all duplexes as well.
A request for "standard cables" has been sent to John Elias, who forwarded it to Keith Schuh and JJ.
Rob will order sturdier industrial shelving. It should arrive by Friday.
While the SCPU's are being moved to the 1st floor, some PC's will be moved to the shelves, and technicians will install additional hardware for cabling.

Jeff Tseng / MIT / jtseng@fnal.gov