Minutes SLAC-ALTAS TDAQ Meeting 29-Jun-2006
-------------------------------------------

1. Round table reports:

Stephen:

- Setting up a new "project" with packages that can be used both
  online and offline.
- CVS project is set up. Figuring out how to build the packages.
- All packages already in existing CVS trees, won't move them.
- A project is essentially a bunch of cmt files controlling which
  versions of which packages to include, no extra code.
- Name shortened: "DetectorCommon" -> "DetCommon".
- Other subsystems? - One of them was intersted, don't recall which
  one just now.
- Should know how to build things by the next meeting.
- Make up tests for this and for HLT.
- No drop-dead time but would be nice to have this as soon as
  possible.  Would be useful to have this in 12.0.3 (1st August).

Discussion which release to import:

- Stephen suggests 11.5.0 which is used in production right now.  But
  doesn't contain the full Trigger. Some Trigger people using the
  nightly.
- Want to pick a stable release but also one that has what we need.
- Importing parallel releases is straight-forward, takes ~1 day.
- New tracker got first serious tuning in 12-series.  So would want
  that from b-tagging perspective.

Amedeo:

- Changed the way the authorization is made between proxy and server
  and client and proxy.  Previously proxy was passive.
- Passive doesn't work because if client doesn't make the
  authorization soon enough will drop out.
- So now proxy has to know the password and the protocol.
- Server sends back 20 bytes that are encrypted (challenged response).
- Have to use MySQL code to do this, so introduces a dependency.
- Now introducing a control pass, so we can independently control the
  proxy, e.g., tell it to flush its cache, etc.  Expect to have this
  working by end of the week.
- Multicast may be working by next week.
- You have a request and a result set.  The entity that pushes will
  encapsulate them and prepend with a protocol header.  Will decide
  whether to give the ability to push to a proxy or a separate entity.
  For now put this into the proxy.

Discussion on packet loss:

- Steffen suggests for the multicast setup it would be very useful not
  to have to do that statically.  Works for O(10) nodes but not clear
  for O(1000).  Possible race condition after joining a multicast
  group may lead to packet drop.  Better not make too many reliability
  assumptions.  In particular, if data is spread over two or even man
  packets.  Drop probablity could be as high as 10e-4(?) per event
  (for ~20 packets).  But the bigger you go the more likely to fill up
  buffers.  For configuration data of O(30) MB we would have thousands
  of packets.  Also, sender assumes homogeneous system in receiver
  capability.  Bottom line is that packet drop may become more of an
  issue with dynamic multicast.
- 50 nodes should be a very good test.

Steffen:

- Test farm: end of BaBar data taking dalayed to August 21, so test
  farm will be set up in (early) September now. SCSS is informed.
- No progress on the infamous Oracle server.  Will ask Chuck.
- Stackable switch about $7k. Has routing capabilities, so can be
  isolated from other networks, important for doing multicast tests
  (without bringing the lab down).
- Prefer SLAC standard over the switch that CERN is going to buy
  towards the SCCS side.  Could add a CERN switch behind it.
- In principle want the exact CERN switch to make sure the SLAC tests
  are relevant.
- Time scale: would have to buy our switch essentially now.
 

2. A.O.B.

September TDAQ Week:

- Stephen, Su Dong, Rainer, will be going, Amedeo probably, Steffen
  probably not, will assemble the switch.  Need to file travel
  requests soon.


Action items:

- Stephen find out if 12.0.1 is ok or if it's not what's wrong with
  it.  Clarify both from Trigger and Tracking perspective.  Will talk
  to Markus and perhaps Jamie.
- Amedeo ask CERN about which switch they are going to buy and when.
- Steffen go ahead get the purchase order ready for the Cisco 3750,
  hold off until we hear back from CERN.  If they buy the same, we go
  ahead, if they buy something else in >6 months we go ahead, if they
  buy something else ~now, we reconsider.