[tahoe-dev] Precise Puppy (linux) tahoe-lafs 1.10.0 initial report
Ed Kapitein
ed at kapitein.org
Sun Oct 6 09:23:30 UTC 2013
I am running one node of the testgrid on a pi, without any problems.
One of the things that might kill your pi is the oomkiller.
I had numerous ocasions where the oomkiller did not kill a job, but
froze the pi itself.
adding the following lines to /etc/sysctl.conf
vm.overcommit_memory = 2
vm.overcommit_ratio = 80
made the crashes go away.
Just my 0.02 BTC
Kind regards,
Ed
On 10/06/13 04:07, Garonda Rodian wrote:
> Very interesting - your grid looks to have been up for 2 days now :).
> I had investigated the BeagleBone Blacks, but eventually decided that
> on paper the Raspberry Pi was a better very low end platform (2 USB
> host ports vs. 1 was more important to me than the BBB's better CPU
> and lower power draw) and was a bit cheaper, and the ODROID-U2 was a
> better low end platform (quad core ARM + 2GB RAM, still 2 USB host
> ports, still 100Mbps Ethernet but at least it's not on the USB bus
> anymore).
>
> Did you check the Raspberry Pi's on-board voltage at the time of the
> brownouts/crashes, or can you help with with whatever test case
> crashed it, so I can check on that myself?
>
>
> ------------------------------------------------------------------------
> From: jason.johnson at p7n.net
> To: deepside at hotmail.com; anders.genell at gmail.com
> CC: tahoe-dev at tahoe-lafs.org
> Subject: RE: [tahoe-dev] Precise Puppy (linux) tahoe-lafs 1.10.0
> initial report
> Date: Fri, 4 Oct 2013 23:23:20 -0700
>
> I have been using BeagleBone Blacks with debian wheezy. So far so good
> the grid is located https://tahoe.netgreen.us there is 8 bbb in this
> grid. PI seemed to brown out or crash on me during testing. Just
> thought I would toss this in incase you wanted to take a look.
>
> Jason
>
> *From:*tahoe-dev-bounces at tahoe-lafs.org
> [mailto:tahoe-dev-bounces at tahoe-lafs.org] *On Behalf Of *Garonda Rodian
> *Sent:* Friday, October 4, 2013 9:11 PM
> *To:* Anders Genell
> *Cc:* tahoe-dev at tahoe-lafs.org
> *Subject:* Re: [tahoe-dev] Precise Puppy (linux) tahoe-lafs 1.10.0
> initial report
>
> Thank you for the feedback, Anders!
>
> I'm definitely not starting X on the Pi (model B), though I had not
> yet lowered the GPU RAM to 16MB.
>
> Can you give me an idea of what "a bunch of very large files" means -
> 100x100MB? 50x10GB? 4x1TB? I can say I almost always use Sandisk
> Ultra or Extreme SD cards, though I was honestly planning on having
> the storage be on a USB flash drive, leaving the SD 100% for the boot
> drive and OS.
>
> I was actually hoping to run two storage nodes, with either two USB
> flash drives on the same grid, or one USB flash drive and one USB hard
> drive, each on different grids (a "small storage" grid and a "large
> storage" grid).
>
> Back to the original topic, Precise Puppy 5.7.1 on a physical box,
> quad core i7 with 4GB of RAM has now successfully completed one test,
> 100% local, with two storage nodes, one Introducer, and on
> client/Gateway, once I figured out which ports in the config files are
> used for what. Up to a 1GB file was uploaded without a problem,
> though it appears that the bottleneck was the gateway with a CPU
> bottleneck. Regrettably, it looks like profiling the Python code will
> require altering the python code, so I've got to figure out how to do
> that so I can see where the slow point is.
>
> Does anyone know if it's OK on Debian/Ubuntu to statically assign
> ports from the IANA dynamic port range of 49152 to 65535 if the system
> is also likely to assign some dynamic ports? I'm a big fan of knowing
> what your ports are, and that'll be critical once I toss a firewall or
> two into the mix.
>
> ------------------------------------------------------------------------
>
> CC: zookog at gmail.com <mailto:zookog at gmail.com>;
> tahoe-dev at tahoe-lafs.org <mailto:tahoe-dev at tahoe-lafs.org>
> From: anders.genell at gmail.com <mailto:anders.genell at gmail.com>
> Subject: Re: [tahoe-dev] Precise Puppy (linux) tahoe-lafs 1.10.0
> initial report
> Date: Fri, 4 Oct 2013 06:43:55 +0200
> To: deepside at hotmail.com <mailto:deepside at hotmail.com>
>
> Hi again, sorry for replying so late...
>
> The Pis used for storage nodes are in general not used for anything
> else, and we try to keep X turned off to save resources. You can also
> lower the amount of RAM reserved for the GPU to a minimal 16 Mb in the
> config.txt file in the Pi boot partition.
>
> Also, some of our Pis krashed when stressed, e.g. by uploading a bunch
> of very large files, until the SD card was replaced by a
> Sandisk SDSDX-016G-X46. The Pi is notoriously sensitive about what
> card is being used.
>
> Finally, if lack of memory is limiting performance, it is possible to
> set up a swap partition on the Pi. It will slow things down horribly,
> of course, but may just get the job done.
>
> Regards,
>
> Anders
>
>
> 28 sep 2013 kl. 05:20 skrev Garonda Rodian <deepside at hotmail.com
> <mailto:deepside at hotmail.com>>:
>
>
> Thank you for the report on the Raspberry Pi being used in
> production - are you and your friends running just one storage
> node on the Pi, or are you also running any other software (second
> storage node, Tor, I2P, OpenVPN?). My RPi consistently simply
> dies during the trial - no errors, it just... stops, but based on
> your feedback, I'll continue.
>
> As I'm hoping to run some medium scale tests, I'm going to have to
> have something to generate a lot of nodes all at once, and I hate
> wasting effort. At this point, I'm targetting something more like
> the old terminal/3270/DOS menus and/or wizards - simple
> walkthroughs with questions to answer that can be used to create
> the files for an entire grid, or add to an existing grid's files,
> hopefully with some manner of "wrapper" (Tor, I2P, OpenVPN)
> capabilities available as well.
>
> Does anyone have a good Python tutorial for experienced
> programmers? My C and assembly used to be pretty good and my SQL
> is excellent, but I haven't picked up a new language in a long
> time, and I never dealt with parallelization much.
>
> P.S. the Precise Puppy 5.7.1 VM at 768MB fails with the GUI, but
> succeeds at the command line with everything nonessential (cups
> printer daemon) disabled, so the critical memory limit for the
> trial is very close to there, OS overhead included.
>
> > From: anders.genell at gmail.com <mailto:anders.genell at gmail.com>
> > Date: Thu, 26 Sep 2013 19:54:44 +0200
> > To: zookog at gmail.com <mailto:zookog at gmail.com>
> > CC: tahoe-dev at tahoe-lafs.org <mailto:tahoe-dev at tahoe-lafs.org>
> > Subject: Re: [tahoe-dev] Precise Puppy (linux) tahoe-lafs 1.10.0
> initial report
> >
> >
> > >
> > >> P.S. If I'm lucky, the Raspberry Pi has completed its trial
> run, though if this is the RAM requirement, I'm not holding out
> much hope.
> > >
> > > It is too bad about #1476, because I really like to be able to run
> > > unit tests everywhere and all the time. However, I believe
> that the
> > > gateway or storage-server itself will run fine on Raspberry
> Pi, even
> > > if (due to #1476) the tests will fail.
> > >
> > >
> >
> > Just to chime in: We have several storage nodes running off of
> RPis in our friendnet, and they seem to work fine as such.
> >
> > We would absolutely love a setup menu - many of our
> participating friends have never used a terminal. Looking forward
> to be dazzled!!
> >
> > Regards,
> > Anders
> > _______________________________________________
> > tahoe-dev mailing list
> > tahoe-dev at tahoe-lafs.org <mailto:tahoe-dev at tahoe-lafs.org>
> > https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
>
>
>
> _______________________________________________
> tahoe-dev mailing list
> tahoe-dev at tahoe-lafs.org
> https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tahoe-lafs.org/pipermail/tahoe-dev/attachments/20131006/643d39cb/attachment-0001.html>
More information about the tahoe-dev
mailing list