18 Jan 2022
Now that the “Dumpster Drobo” (an old Drobo b800fs) isn’t making
noises like an angry garbage disposal whenever it’s turned on, I
figured it was time to actually play with it. A NAS without any data
on it isn’t very fun.
So I logged onto it from my desktop Mac via SMB/CIFS, and started
copying files to it. And that’s when I realized that the unit might
have been left next to a dumpster behind an office building for a
reason. It was horrendously slow.
As it turns out, the Drobo b800fs has something of a reputation for
being very slow, even among Drobo models – which, as a class, are not
known for speed.
Even after enabling Jumbo Frames (which increases the MTU size to 9000
bytes), the best throughput I could get – no matter if I connected
via SMB/CIFS or AFP – was about 30MB/s write, and 35MB/s read.
This is pretty dismal from an 8-bay NAS running on Gigabit Ethernet,
and I’m at a bit of a loss to explain why. The b800fs hardware
doesn’t look terrible. Although Drobo’s official spec sheets don’t
get into details, it appears that the b800fs hardware is very similar
to the Drobo Pro FS, which was reverse engineered and described in this
blog post by Daniel Parnell. And it comes with a FW800 port!
Looking at my unit compared to the photos of Parnell’s Pro FS
(attached to the blog post linked earlier), one of the only
differences I can see is the lack of a FireWire port right next to an
empty BGA IC site on mine, which I assume is where the FireWire
800 controller would be populated on a Pro. (This makes sense, as
FireWire controllers were notoriously expensive due to licensing fees,
one of the reasons the connection never took off like USB did.)
This makes me suspect strongly that, if I wanted to reopen the case
and start poking around, it would probably be possible to connect via
serial console to either the Linux or VxWorks operating systems
running on the device. And if I can connect to the embedded OS, maybe
I can figure out why the SMB transfer rates are so darn slow. Who
knows, maybe I can even get
rsync working on it; that would
instantly make it a whole lot more useful, even at only 35MB/s. It’s
a bit of a long shot, and I suspect the Dumpster Drobo will end up
back at the electronics recycler eventually, but it seems worth
playing with before I give up and toss it.
First, however, it makes sense to check the USB port on the back of
the b800fs, and see what if anything I can do through it. Hopefully
it’s useful for something; it’d be a bit weird to have removed the
FireWire port and controller, but left a USB 2.0 port on the
motherboard without allowing it to be used to communicate with the
operating system, access the firmware partition, or some other
Read more »
17 Jan 2022
One of my minor hobbies is that I am an inveterate dumpster-diver.
Particularly in the suburbs around DC, where defense contractors pop
up and disappear with the regularity of dandelions, you can find some
pretty great stuff in dumpsters, or just sitting next to them, behind
A few years ago I scored what I thought was a great find: a Drobo
b800fs “storage appliance”, just leaning up against a dumpster
alongside some ripped-out Ethernet cables and CRT monitors.
Naturally, I grabbed it and took it home.
It took me a while to find enough drives to populate it, and then
longer to download the “Drobo Dashboard” software and get everything
working on my main desktop computer. But eventually, I had it all
working. And then… it started making some really awful sounds.
Based on previous experience with old junk, I knew one of the cooling
fan bearings had probably failed and needed replacement.
Although there are plenty of guides around for opening up and
repairing other Drobo models, I couldn’t find much about the b800fs in
particular. So for the benefit of anyone else who may find themselves
with one of these oddball machines, here are some notes on how I
replaced the fans. Hopefully it will be of use to someone, someday.
Taking it apart
First, shut the unit down remotely from the Drobo Dashboard utility. Then,
after it has time to completely turn off and spin down, unplug the
Ethernet and power cords. It’s quite heavy when fully loaded with
drives. Set it on a work surface and remove the front cover, which is
held on with magnets (clever, really). If you remove the drives, mark
them carefully so you can put them back into the same positions. (I
did not remove them, because I like to live dangerously.)
After turning it upside down, remove the 4 small Philips head screws
nearest the outer corners. You do not need to remove the rubber feet,
as you apparently do with some other Drobo models. Turn it back over
to normal orientation, and slide the metal case off the front. There
is a rubber gasket around the front face that may or may not come
along for the ride.
To access the fans, remove 4 more small Philips-head screws on the top
of the exposed sheet metal chassis, near the rear. Gently lifting
this up (note the wires connected to the components on the bottom of
it), you can see the motherboard, fans, battery, and other internals.
Mounted on the back of the now-loose part is the power supply.
Carefully disconnecting the two Molex connectors will allow it to be
completely detached. The power supply contains a number of fairly
large, presumably still-charged, capacitors, so use some reasonable
degree of caution and don’t stick a screwdriver in there.
The battery, which is zip-tied to the side of the chassis interior,
appears to be a standard item, and supposedly keeps data in the write
cache memory alive during a power outage. I noted the specs for
future replacement, but didn’t bother ordering one at this point. It
3.7 V, Li-ion, 7500 mAh
Cell from Japan
This appears to be the same battery used in the Drobo b800i, which
makes sense, since the b800i and b800fs are apparently the same
hardware but with different software (the “fs” version is a
traditional NAS, while the “i” version exposes itself via iSCSI).
The plastic cover on the rear can be removed by reaching into the case
and gently retracting the 5 plastic tabs that hold it on: there is one
in each corner, plus one in the center of the top rear.
The two cooling fans are held in place by “plastic rivets”, which can
be removed by pushing a thin metal tool, such as a nail or small
screwdriver, through the hole in each corner of the fan, and pushing
on the center of the rivet. They will pop right out the back.
Alternately, you can work them out from the exterior of the case by
working a thin prying tool under the rivet “head” and pulling it out.
Be careful not to lose any of the rivet parts; each one consists of
The noisy fan in my unit is labeled “ADDA DC Brushless Model
AD0912UB-A70GL, DC 12 V, 0.30 A, 14P3” and has two wires, black and
brown, connecting it to the mainboard. The connector is a small 3-pin
connector, of which only two pins are used.
Based on a quick search, it appears the AD0912UB-A70GL (what a
mouthful) is actually a pretty common item, with an exact replacement
selling for under $15 at Newark and under $25 on Amazon. Per
Octopart, it’s described as: 92x92x25mm, 12 VDC, 60 CFM, Ball Bearing,
Ultra High Speed Fan. Newark gives us more specific specs: 56.6 CFM,
3300 RPM, 3.48 W, noise rating of 39.4 dBA. Filtering Newark’s
catalog for 92mm square fans operating on 12V with 2 lead connectors
gives some replacement candidates.
Although it’s possible to purchase the exact OEM part, the Bisonic
BP922512H-W is about the same price, pushes more air (58.7 CFM), runs
a bit slower (3000 RPM), and has a noise rating that’s slightly better
(34 dBA). The power consumption is just a tad more at 3.94 W according
to the datasheet. Looks good. I ordered two of them, figuring that
if one fan had already failed, the remaining one might not be too far
behind. Might as well do them both at once.
Swapping the parts
After waiting a few weeks for Newark to deliver the new fans, it was
time to install them in place of the old ones. Because the new fans
come with bare wire leads, some basic soldering is necessary to attach
them to the small 3-pin connectors used to connect to the mainboard.
Rather than cutting and stripping the existing wires on the old fans,
I noticed that the solder points attaching them to the fan PCB were
easily accessible by peeling back the label in the center of the fan.
A bit of heat applied to each solder blob got them loose. With that
done, it was a simple matter to splice the wires together, and cover
the joints with some small heat-shrink tubes. Black goes to black,
and brown goes to red on the new fans.
Reassembly was straightforward. The new fans line up exactly with the
existing chassis holes, and the existing plastic rivets seem to hold
them firmly in place. After closing up the case and connecting power,
both new fans spun up immediately at the beginning of the boot
process. After a long weekend of “burn in” time (basically waiting to
copy any data to the device in case one or both fans failed quickly),
I’m ready to declare this repair project a success.
Read more »
03 Jan 2022
In “The computer built to last 50 years”, Lionel Dricot
(aka ploum) makes a compelling case for a new type of personal
computing device, one designed to last decades and “do timeless
activities and do them well”. Altough I have some nits to pick
regarding the specifics of how a machine like this might be built and
what it might include, the concept itself is a great starting point
Rather than looking only at the next incremental improvement in
hardware and software, the question of building a computer that will
still be useful in five decades forces us to think backwards from
high-level, “timeless” use cases that address fundamental human
problems. What do people use computers for? Communication,
primarily. But also accessing reference information. Writing
documents. Reading books and periodicals.
None of these tasks are especially computationally intensive, at least
by modern standards. (Sure, there was a time not that long ago, when
justifying variable-width text or automatically using ligatures and
hinting was considered a reasonably hard problem. Progress moves, but
slowly.) The most complex number-crunching a purely text-based
machine like this would need to perform would likely be cryptographic
calculations: support for modern crypto is a must, in order to
interact with modern networks and speak protocols like TLS.
At the risk of a bit of a derail, this is basically the limiting
factor on whether a computer is still useful today. If it can run
OpenSSL, GnuTLS, or another modern crypto library, you can probably
hook it up to the Internet and do networked stuff with it. If it
can’t, you can’t – and you’re generally limited to using it in
offline roles, or sticking it behind some more modern system and
using it as a thin client, which is clever but also cheating.
This results, incidentally, in a sort of floor separating the
still-potentially-useful hardware – that which is capable of running
modern software – from the true “retro” antiques, which cannot and
probably never will. (From what I can tell, the oldest production
computer that is still supported by Linux is likely the IBM S/390,
released 1990. So if you happen to own one of those, you’re already
three-fifths of the way to a 50-year computer!)
Provided there are no game-changing revelations in cryptography in the
near future, I expect this dividing line to be largely stable.
(Sudden developments in quantum computing could, in the worst case,
make anything non-quantum-capable into insecure antiques overnight.)
I think we should be able to build a 50-year computer with hardware
available on the market today, if it were designed well.
Of course, that “if” is pulling a lot of weight. But that’s the fun
of the challenge.
Read more »