Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009 PSI-XFEL Challenges and Developments.

8
Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009 PSI-XFEL Challenges and Developments

Transcript of Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009 PSI-XFEL Challenges and Developments.

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFEL

Challenges and Developments

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFELPSI-XFEL

XFEL Project Schedule

250 MeV Test Injectoruntil fall 2010 PSI-XFEL

construction startspring 2011

Gun Testuntil summer 2008

4 MeV Testuntil spring 2009

continued

continued

?

?

■Have to support 3 machines at the same time!►Lots of diagnostics►Frequent modifications

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFELPSI-XFEL

250 MeV Injector Construction Site

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFELPSI-XFEL

Control Room

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFELPSI-XFEL

250 MeV Injector in Numbers

■ 48 VME IOCs►RF IOCs: 21►Magnet IOCs: 2►Timing IOC: 1►Diagnostics IOCs: 24

■ 15 PC IOCs►Vacuum IOC: 1►Diagnostics IOCs: 4 + 10 camera servers

■We expect a factor of 10 for "the real thing"

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFELPSI-XFEL

Networking

■ Class A subnets preferred►maintainability issues► fear of broadcast traffic

■ But 500 IOCs do not fit into one subnet■ How to split the network?►By system? (RF, magnets, diagnostics, …)►By sector? (gun, buncher, linac, …)►Where to put the control room?

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFELPSI-XFEL

backbone network(control room, central IOCs)

Network Layout

...

...

EPICS

non EPICS

gun linac 1 linac n undulatorsbeamline 1

beamline nvacuum system PLCs

machine interlock system PLCs

web cameras

VLA

N r

ou

ter

...

Does this make sense?

CAGW CAGW CAGW CAGWCAGW

CAGW

Dirk Zimoch, EPICS Collaboration Meeting, Vancouver 2009

PSI-XFELPSI-XFEL

Gateway Reliability

■ Runs stable at SLS►16 beamline networks (up to 3000 channels)► readonly office network gateway (up to 40000 channels)

■ Virtual machines (vmWare) on high reliability blade system■ Found problems with waveforms►10 records x 2000 doubles x 10Hz►Waveforms came though but all scalars were dropped►Solution: increase vxWorks TCP queue size! (8k → 64k)► Is the gateway too slow reading waveforms?