1) Paper
the TR version of the paper is done... we're just waiting for a name we
agree on. No word from the IEEE Micro folks
2) New Data
We will likely be getting some data from ANS's surveyor project to
calibrate John's measurements (courtesy mattz, previously of Midway
fame). They have 30 sites doing frequent delay/loss measurements
between each other. Better still every measurement is timestamped off a
GPS with 50us accuracy, so it'll be possible to use the data to do
simultaneous measurements. UW is running a surveyor node. We agree to
only use for research, not to redistribute, and to thank them profusely
in papers.
3) New machines/cluster room
The cluster is being moved... new subnet space (128.95.219?) has been
allocated and new power is being added. Porky 2 has received 20 new
machines. We will be ordering some of these two (I've placed and order
for 6 to start). Two will be shipped to UCB and UT, one for nimi and
the other three are for Access/Detour stuff. Extra FYI: We have
requested a PRIVATE subnet from cac so we can isolate Detour from the
dept and do our own subnetting within the block.
4) Remote Boot for Access
Machines supporting Intel's Wired for Management standard (such as the
new Dell machines) have a standard BOOT ROM on their Ethernet cards
(PXE). David and Eric have downloaded a public domain program that uses
this PROM to load complete environments onto machines. They've done it
successfully for linux, david is looking at Win95 and NT, and Eric is
working FreeBSD. One problem is that the utility actually understands
filesystems (FAT and ext2) and doesn't write NTFS or ufs. We talked
about the possibility of writing raw blocks assuming that the total
image size is less than the partition size. This is being investigated.
Also, we need to understand what needs to be done to
configure/specialize these boxes after a generic image is written. This
seems easy for linux/Free, butmay require some more knowledge for NT.
5) Detour status
Eric says the user level framework works well and basic routing (LSA
exchange, updates, etc...). He next needs to work on the actual routing
algorithms. Andy reports that the detour kernel basically works
although NAT is still crashing sometimes.
6) Quals
Neal and Andy will do quals next week.
7) Misc
We talked a bit about statistical sampling methods basic on the bellcore
and queso web pages sent out yesterday. The technique seems to have
broad applicability for getting a feel for host trends.
Also, early indications from the web trace data reportedly indicated
that real audio is the largest traffic class (in bytes) entering UW.