[tahoe-dev] dogfood tasting -- robustness of the volunteer grid
zooko
zooko at zooko.com
Thu Apr 9 12:23:32 PDT 2009
Today there was a major Internet outage in California. This means
that the mail server for the tahoe-dev mailing list is unreachable,
which means you won't read this until after the problem has been
repaired, and by then you may well know more about what happened than
I know now.
One interesting during this process was that I got to see how well
three different Tahoe grids stayed up and provided service during the
outage. The first is the allmydata.com grid. Alas, all of the
allmydata.com storage servers were cut off from the Internet, or at
least unreachable from my home here in Boulder, Colorado. Also the
http://allmydata.com web site and the Tahoe webapi server, and the
FTP server, etc. Allmydata.com currently uses Tahoe to store
customer files, and the massive robustness of Tahoe means that those
files are likely to survive accidents -- they have good "durability"
-- but the current network configuration means that they are not
particularly more likely to be reachable during Internet outages --
they have pretty normal "availability". (Actually their availability
is probably a bit better than most services, since disruptions which
effect only one or a few of their storage servers do *not* reduce
availability of their service to their customers.)
(Note: I am no longer an employee of allmydata.com and don't speak
for them. I am commenting on my observations or guesses about their
situation. Peter or Zandr could correct me if I am wrong.)
The next grid I looked at is the test grid. This is running mostly
on servers contributed by and operated by allmydata.com, plus whoever
else boots up a Tahoe node and connects to it to experiment. I
couldn't connect to any of the servers on the test grid because the
test grid introducer itself is inside the network outage area. I'm
pretty sure that even if I could reach the introducer, that all of
the files on the testgrid would be unreachable due to so many
testgrid servers being inside the outage area.
The next grid I looked at is the volunteer grid. I was at at
coffeeshop in Boulder when I started wondering about this, so I tried
to connect my laptop to the volunteer grid. Alas -- the volunteer
grid introducer is running on my server -- nooxie.zooko.com -- which
is in the same giant co-lo building in San Francisco as the
allmydata.com servers. This made me wish for decentralized
introduction. Fortunately, there are some students interested in
hacking on decentralized introduction this summer, as prompted by the
Google Summer of Code project. More on that later.
Then I went back to my house, where I have a volunteer grid node
already running. Hooray! It doesn't need the introducer, since it
already knows the furls of all of the (current) volunteer grid
servers. It looks something like this (attached as grid-status.html):
Nickname Connected? Since
SECORP-MAC No 2009-04-09_10:58:20
SECORP_DOT_NET_02 Yes: to 72.14.177.254:41267 2009-04-08_07:24:33
stockrt-terra No 2009-04-09_10:58:20.338400 2009-04-07_16:47:54
draco (Zooko's Mac/PPC 867 MHz laptop) No 2009-04-09_09:25:43
trid0 Yes: to seron.dyndns.org:3141 2009-04-08_07:30:41
SECORP_DOT_NET_04 Yes: to 72.14.177.254:41170 2009-04-08_07:24:33
SECORP_DOT_NET_03 Yes: to 72.14.177.254:35056 2009-04-08_07:24:33
yukyuk (Zooko's Athlon64 Linux workstation) Yes: to (loopback)
2009-04-07_16:47:54
francois1 at tahoe.ctrlaltdel.ch Yes: to 62.220.138.72:50029
2009-04-08_07:31:23
SECORP_DOT_NET_01 Yes: to 72.14.177.254:44793 2009-04-08_07:24:33
ndurner Yes: to ent.merseine.nu:4961 2009-04-08_07:27:37.321217
2009-04-07_16:47:54
trelbox Yes: to trel.dyndns.org:56034 2009-04-08_23:26:26.063565
2009-04-08_22:54:07
aogail-volunteergrid @ tigard.w007.org No
2009-04-09_10:58:20.347164 2009-04-08_17:13:52
I ran a deep check of my directory of flac files, and here are the
results, below, and attached as deep-check-flax.html. Now that's
more like it -- all of my files are still usable. Just for kicks, I
played some music for myself from the volunteer grid.
Once we have a map with little colored pins in showing the locations
of the volunteer grid servers, I will be able to visually imagine
what sorts of internet outages the volunteergrid can tolerate without
interrupting operation.
Hm... I wonder if we could use one of those service that tries to
map your IP address to a geographic location to automatically
generate such a map.
Regards,
Zooko
Relative Path Healthy Recoverable Storage Index Summary
<root> False True tsxqypdes5ysfme6cuqebinmdy Unhealthy: 8 shares
(enc 6-of-9)
01 Battery.flac False True efqq2yffg36yvclyd5wff5wwrm Not
Healthy: 5 shares (enc 4-of-6)
02 Master Of Puppets.flac True True inal62aoisoa4huma5fmzq3req
Healthy
03 The Thing That Should Not Be.flac False True
4ub2wplyahagoner4npnzbshle Not Healthy: 5 shares (enc 4-of-6)
04 Welcome Home (Sanitarium).flac False True
jjrflgi4bpkodj5q7ygz6wz6oy Not Healthy: 5 shares (enc 4-of-6)
05 Disposable Heroes.flac False True tdsby2jfikwsy67vfgoomax26e
Not Healthy: 5 shares (enc 4-of-6)
06 Leper Messiah.flac False True figj73pp7e7swislhr7bo3ol2u Not
Healthy: 5 shares (enc 4-of-6)
07 Orion.flac False True crcijgwqs7wmzd4juhcu4xpy5u Not Healthy:
5 shares (enc 4-of-6)
08 Damage Inc..flac False True tydsprhi2bkdnqhzxsdzpgbjdq Not
Healthy: 5 shares (enc 4-of-6)
13-of-16-fec False True xzazzqtgytts6mp3iiv6tjrxdy Unhealthy: 14
shares (enc 13-of-16)
13-of-16-fec/01 Battery.flac False True wdbbrh2y4p77y7jpflkbucuzvy
Not Healthy: 15 shares (enc 13-of-16)
13-of-26-fec False True igmsz64hn6egnqu5c6e4bjmtsq Unhealthy: 23
shares (enc 13-of-26)
13-of-26-fec/01 Battery.flac False True uwgvqn372sycxncb6p3ct6rd7i
Not Healthy: 23 shares (enc 13-of-26)
13-of-26-fec/02 Master Of Puppets.flac False True
uwqr6zia2qn2sgd2mxbwzui6ei Not Healthy: 23 shares (enc 13-of-26)
13-of-26-fec/03 The Thing That Should Not Be.flac False True
xollsyamtkwcg3nxhxwl4qffyq Not Healthy: 23 shares (enc 13-of-26)
13-of-26-fec/04 Welcome Home (Sanitarium).flac False True
2izwujkldbas3z2oryunctseoi Not Healthy: 23 shares (enc 13-of-26)
More information about the tahoe-dev
mailing list