I reported this bug few times already, with different versions of BOINC.
Each time it was considered as non-replicable and ignored.
Well, will try third time.
Host OS: Vista x86 SP1
BOINC 6.6.36

Network config:
host with BOINC connected to internet via another host in local network with NAT (as native Windows XP internet connection sharing) enabled.
When host with Windows XP lost its VPN connection, i.e. lost internet access,
BOINC's daemon on Vista host hung badly. BOINC manager tried to connect to it but failed each time.
OS/BOINC restarts didn't help.
Only when VPN connection was restored and all hosts in local network recived their internet connection back BOINC manager was able to connect to BOINC daemon process.

Again, I've seen such behavior before and it's still with us in BOINC 6.6.36.

---

Corresponding ticket added:
http://boinc.berkeley.edu/trac/ticket/938

---

I'm wondering whether to add a cross-reference to BOINC cc using excess CPU as service, which bit me again today. Internet access was available, but the problems I've documented in that thread tend to occur when Microsft Automatic Update has downloaded patches but not yet been given permission to install them - as, obviously, today.

Note. My observations relate to BOINC v5.10 - but before you patronise me :-), today's was on a MS Small Business Server, which has to be a Domain Controller. So no v6 version can be installed.

Similar observations - BM can't connect to the daemon, even though the daemon is running (and using excess CPU), according to Task Manager.

And because it's v5, there is an additional test I can perform which Raistmer can't. If I stop the service using the service manager snap-in, I can start BOINC in user mode (v6 just re-starts the daemon). In user mode, the exact self-same boinc installation uses zero CPU (unless asked to do something, like a manual project update), and the Manager is responsive and fully functional. Close the Manager, restart the service, and the Manager stalls at start-up (with two processes showing in Task Manager). Kill one process, and they both stop, and BM closes.

So there is something specific to running in daemon mode in all this.

Some addon:
In my last case BOINC runs not as service (can't install as service on Vista cause need to run CUDA tasks).
But the same situation was on Win2003 x64 Server where BOINC installed as service.
The problem is with boinc.exe process that refuse to communicate (and uses excessive CPU amount for itself indeed) with boincmgr.exe

---

This is a hint to the developers to do some more testing on this. It is easily reproduced.

Systems Win X64 BOINC 6.6.36 and 6.6.37.

I got 3 systems running all with Seti and about 20-50 Uploads waiting each.
My own Internet went down. I did a reboot on 2 systems.
And then... BOINC manager could not connect and when I looked at taskmanager, I saw almost no BOINC activity but a lot on VSSERV.exe almost 100%.
The tasks (10 of them) where running 1 minute, where closed (disappeared from taskmanager and the computer was near 0%) then after some time 100% running of the 10 tasks.
This was going on for 45 minutes on 2 computers.
I got a short glimpse of the log file and I saw a lot of can't resolve hostname errors. But as I could not properly connect, that's all I could see.

I made my own program to connect to BOINC without the manager and what I saw then was that the tasks were indeed running. But only at a pace of 15 seconds every 5 minutes or so. Te elapsed time was moving at this rate.

One of the system is dual boot, so I booted on Win 7 with no uploads pending, no problems whatsoever.

My guess is that because of the lack of an Internet connection on my side BOINC keeps trying to connect and get a DNS resolve. And this causes a lockdown.

Once my Internet was working again, everything went back to normal.

So a lot of uploads pending and no Internet = hangs.