Raspberry Pi 3 vs Cubieboard 2 Samba NAS performance

I’ve been using a Cubieboard overclocked to 1.3GHz as my home little NAS server since a few years ago, and just recently I bought a Raspberry Pi 3 thinking it would have been surely faster.

I was especially looking to get better Samba performance.

I have additional ntfs-3g and encryption layers other than Samba, but both boards had their version of Debian installed, with the same packages and same configurations, so the starting configuration was basically the same.

Both were -alternatively- attached to the same Toshiba Canvio 3TB USB3 disk (working in USB2 mode on both boards), and on two adiacent ports on my TP-Link ADSL router.

I overclocked my Raspi3 to 1.5GHz, with sdram_freq set to 500.

For my test, both were alternatively running off the same AC powered USB adapter, under the same wattmeter, each with his own USB power cord.

Well, it was an overall disappointing experience.

Raspberry Pi 3 @ 1.5GHz

  • Download speed via Samba: almost 7MB/s
  • CPU temperature: about 42°C
  • System load: ~2
  • Power draw: 2.6W, with short spikes to 2.7W

Cubieboard 2 @ 1.3GHz

  • Download speed via Samba: almost 7MB/s in average, oscillating between 6.3 and 7.5MB/s
  • CPU temperature: about 42°C
  • System load: ~2
  • Power draw: average 3.4W, ranging from 3.3 to 3.5W

So, the only area where raspi3 is superior to cubie2 is the power draw, otherwise I would have expected way better performance.

Interesting to note, the transfer speed via samba was the same on the raspi3 at the default frequency of 1.2GHz (no benefit from overclocking whatsoever, if not maybe increased oscillations in speed with the same average value), while cubie2 seriously improved after I overclocked back then.

This might indicate a saturation of the USB bus, where the ethernet adapter resides in the raspberry pi 3, which might be the case since the system load of 2 means the 4 cores of the raspberry pi 3 are loaded half as much as the 2 cores on the cubieboard 2, so they could be waiting for the bus to free itself.

Investigating further, I ran

vmstat 3

on both the boards during the transfer… and averages were as follows:

  • Cubieboard 2: user 7%, system 52%, idle 39%, waiting 2%
  • Raspberry Pi 3: user 2%, system 28%, idle 55%, waiting 15%

I think it’s quite clear at this point that while the cubieboard 2 is crunching during that time (the idle should be due to one of the two cores not being optimized), the raspberry pi 3 is waiting on the bus to free itself before sending in other data (the higher idle also must be due to the 4 cores not being all used for the task, mostly 2 of which just don’t do anything at all).

I spent about half to buy the raspberry pi 3 than what I paid for the cubieboard 2 back then, but after so much time it’s not surprising; what is surprising is that the everyday performance appears to be the same 🙁

At this point I will probably keep the cubieboard 2, because it’s already configured as I want it (it’s got more amenities to it than just the NAS server based on Samba) and the different power draw would save me only about 2€ per year on the raspberry pi 3, so no point in upgrading.


It’s been a couple months since i wrote this article, and for the last month, maybe more, I’ve been actually using the raspi3 as server (and have sold the cubie2 which wasn’t useful anymore).

I noticed that, on the long run, the NAS transfer speed didn’t keep up on the cubieboard after the uptime went, well, up 🙂 It got down to about 2MB/s, with spikes of 5MB/s, maybe less, but never quite reached the 7MB/s average which I got for the tests. No idea what caused this performance drop, but it was a fact: as days went by, the transfer speed was not on par, but far less.

The raspi3 on the other side looks more reliable, its uptime is 40 days as I write, and it still is right below the 7MB/s mark.

Firefox not firing onReadyStateChange with two AJAX requests and a confirm dialogue

There are many discussions like this on the internet, but I am confident I am now examining a very specific problem of Firefox, with a very specific AJAX routine.

Symptoms: there are other AJAX requests in the background, and there is a request initiated by the user, which requires confirmation (your average “are you sure you want to delete?” dialogue). The AJAX request to delete the entry never fired. Every AJAX request was in the default, asynchronous mode.

No matter what, I couldn’t for the life of me make that request work, everything else did. Only the “delete” functions didn’t (in Firefox, as Chome and Internet Explorer worked no problem).

I used console.log messages as crumbs along the way, to follow the flow, and it stopped right at onReadyStateChange, as anything after it wouldn’t show zip in the console. Which meant the onReadyStateChange wasn’t being triggered.

I even thought it was something off with the function name (maybe the delete inside it borked something, silly me thought).

I also tried switching to synchronous mode the incriminated requests… and welp, they worked! For a short while I thought about leaving them be synchronous, but I didn’t want those pesky warnings in the console about synchronous requests ruining the user esperience.

Only by chance an epifany happened, and the obvious came forth (I KNEW there had to be a reasonable explanation which I had yet to grasp).

Well, my configuration went like this (it’s an internal web application we are talking about): an AJAX request triggered each second to a PHP to check the user credentials and inactivity time, and an AJAX request by user intervention was launched to delete an entry… after a confirm dialogue requested, well, user confirmation for the delete (duh).

Everything worked flawlessly before, the problem started appearing after I added the background AJAX request each second (which I didn’t want to get rid of anyway).

As you may or may not know, the window.confirm dialogue is a synchronous structure, which means the page execution is frozen until the user presses either OK or CANCEL in the dialogue (same thing happens with the window.alert function).

So, imagine the background AJAX request being queued behind the confirm dialogue, and obviously taking precedence over anything else that had to come after, and the new AJAX request being formed right after the press on the OK button of the confirm dialogue, to delete the entry that the confirmation was for.

Well, these two requests were launched at the very same time, in a manner that Firefox (not the other browsers though) probably overlapped the onReadyStateChange and never triggered the one for the secondary AJAX request.

In fact getting rid of the confirmation dialogue solved the problem.

Either that, or I will have to come up with an asynchronous method of requesting user confirmation.

Import QR vCard with multiple words name and surname in Android

First of all: YOU CAN’T.

Now for real, there’s a workaround, just buid your vCard text to be rendered as QR Code by removing the spaces between the first name words, and the last name words (surname that you may like).


N:Johns Phillips;John Philip;;; (Last Name;First Name;;;)



and then, in the address book editing form that appears in Android you just tap in the middle of the joined words and press space.

It will be MUCH faster than moving around in the fields the misplaced name components.


There is NO WAY to actually have a vcard data correctly formatted to be correctly rendered by a QR Code scanner and correctly passed to the address book.
If you used a correctly formatted vCard file to directly import it in Android address book, it will work (for example, first export a vCard file for a contact, and then reverse engineer it… but to actually copy the VCF to the phone for each contact and manually importing it would be time consuming and bothersome); but if you use a generated QR code from the vCard code, each and every QR scanner out there will pass the full name data to the address book as a single unformatted string, IGNORING the name components of the N field, leaving the address book parser to decypher which is which.

So, depending on how you go about hacking your way into the vCard format, you might get

John (first name) Philip Johns (middle name) Phillips (last name)

or any other weird combination, BUT the one you were looking for.

Believe me, I tried, and chances are, if you ended up here, you did, too.

In my case I spend on it much more time than I care to admit. More than I needed to actually code from scratch the PHP to generate the relevant QR code into my management software.

So just embrace this workaround, but if you find a solution, please Please PLEASE share in the comments.

Create STL from DICOM with 3D Slicer

Pretty much a self reference post, based on Slicer v. 4.6.

  1. Open the relevant DICOM set.
  2. Open the Crop Volume module
  3. In “Input ROI” dropdown select Create new AnnotationROI
  4. In the slices view adjust the selection so only the interesting area is covered
  5. Click on the Crop! button and see how the slices are cropped to the set area
  6. Open the Editor module (you can select the GenericAnatomyColors palette)
  7. Click on the coloured rectangle on the Label: row, and choose your relevant structure (if you’re here you probably want bone)
  8. Click on ThresholdEffect button (third from last from the effect buttons)
  9. In the Threshold Range input area that just appeared set a starting value of 300 HU (Hounsfield Units) which usually is a good starting density value for bone
  10. You will see the relevant areas blinking in the slices images, and you can adjust the density range by using the small arrow buttons on the side of the lower value of the range
  11. When satisfied, click Apply
  12. Click on the MakeModelEffect button, first from last, and right after click Apply
  13. After a while (depengin on how large is the ROI and how complex is your structure) the rendered model will appear in the 3D view
  14. Click on the SAVE button in the top tools bar, and deselect every entity except the last one (usually named bone if you followed this tutorial), and then choose the STL format in the dropdown on the side (VTK format is selected by default), plus change to your needs the saving folder, and press the Save button
  15. Profit!

Share mysql and apache folders from windows XAMPP to linux XAMPP

This is a replica of the article by jultech from 2007 for posterity and data-replication purposes.

When dualbooting between windows (first OS installed) and linux, do this:

~ # mv /opt/lampp/var/mysql /opt/lampp/var/mysql.BACKUP
 ~ # ln -s /mnt/windisk/path/to/XAMPP/mysql/data/ /opt/lampp/var/mysql
 ~ # mv /opt/lampp/htdocs /opt/lampp/htdocs.BACKUP
 ~ # ln -s /mnt/windisk/path/to/XAMPP/htdocs /opt/lampp/htdocs

This creates a backup of the original XAMPP folders in linux, and them symlinks them to the windows disk mounts.

Restart XAMPP to have th configuration running.

Thank you jultech.

Fix mount: unknown filesystem type ‘isw_raid_member’

You may also get ‘linux_raid_member‘ instead of ‘isw_raid_member‘, usually it’s the same.

So you probably have run through other tutorials telling you not to mount directly but to mount the raid instead…. but you have no raid!

Maybe though, you HAD a raid back in time, and you removed it leaving the disks as separate storage, and in that case there would be “rogue” raid metadata on the disk that makes linux (mint in my case) think it is a raid, and get confused.

Thanks to Chris_F I finally solved it, you need to use dmraid (install it if it’s not on the system, via apt-get) to remove those orphan tags:

sudo dmraid -rE /dev/sdb

In my case, this instantly showed the disk in file manager.

Xiaomi Mi Band 2 vs Teclast H30, first impressions

I had been using the Mi Band 2 foralmost a month, but ordered the Teclast H30 because I liked the SMS text function, and also the general look of it.

What I like of the Mi Band 2 is the fact that I have a watch on my wrist once again (I have been wristwatch-less for about 10 years!), it warns me of incoming calls (when I am working among loud noises or walking to work among traffic, I might not hear my phone or feel it vibrating inside my pocket) and the sleep tracking function.

Heartrate measurement is a disappointment, but not because of Mi Band 2 in itself, more that as a technology you need to stay really in one place and not moving, otherwise it WILL miss hearbeats and give you a really low frequency. Also you do not get continous heartrate measures with the Mi Band 2.

Distance and calories are so useless since they are calculated based on steps count, and amount to a really insignificant number.

When I got the Teclast H30 I put them together on the same wrist. Pairing the band with my phone was a blast (with the Mi Band 2 it took WAY too much fiddling with my phone’s bluetooth).

Surprise, Teclast H30 supports continuous heart rate measurement via a setting on the Android app, but the value was inaccurate if compared to Mi Band 2 (I was fast walking back home, Mi Band 2 gave me a realistic 139bpm, while the H30 was stuck at max 93, but it might as well have been a defect in positioning it on my wrist).

Also Teclast H30 over Mi Band 2 has the find my phone function, where you would jump to said function and your phone would ring.

Also Teclast H30 often was faster than Mi Band 2 in showing the time after raising my wrist.

TSports app doesn’t require online registration as MiFit does.

The advantages though end right about at this point.

I could greenify MiFit app, but TSports is hard to be put in hibernation as it installs as a resident app by default and eats away at the phone battery even if it is in the watchlist of Greenify.

You cannot choose which elements to display on Teclast H30 like you can with the Mi Band 2, you always have to scroll through the Time/Date/Sleep hours/Steps/Distance/Calories/Battery/Find Phone, and it gets annoying really fast! Also on the Mi Band 2 you can display, with the latest firmware, time and date together, while on Teclast H30 they are on different screens.

Finally, the alarms: on MiFit you can set alarms depending on the weekday or just once, while on TSports you can only set an alarm to ring every day for the rest of existence.

I cannot report about sleep tracking nor battery duration as after a half-a-day experience I gave this band to my S.O.

I can testify anyway on the SMS text display to be working, as I got one during that time. It doesn’t display the author name even if it is in your phonebook though, just the originating phone number.

That’s all folks!

SQ8 mini camera DVR, enable loop recording

Connect the SQ8 camera to a USB charger. It cannot be a PC, otherwise it will work as an external drive and show you the contents of the SD, it just needs to receive power.
As soon as you connect the camera you will notice the rear led turning on and cycling through different colors, after which it will turn off: at that moment, the recording will have started.
I still haven’t tested the cyclic recording functionality, but this is supposed to record max 5 minutes chunks of video until the memory runs out, after which it deletes the oldest recordings.

BTW, the correct orientation of the camera is so that the flat side opposite to the buttons side is down, position it elsewhere will give you a rotated video.

On a side note, this camera has a really bad space usage; I put a 32gb card in it, and the file size of the 5 minutes video chunks is 400MB! There is virtually no compression on the video, and the maximum recording time before the oldest files are deleted is less than 6 hours.
90mbs per minute is a really bad stat, in comparison a naruto episode should be 2.4GB if it had the same time/size ratio, it is instead one tenth of that…

Revert, undo & go back from adoptable to portable storage in android marshmallow

I replaced my previous 32GB microsd in my note 3, with a transcend 64GB, and the system asked me if I wanted to use it as extended internal storage. Why not, I never use the sd as swappable storage anyway, and I liked the encrypted storage.

Go ahead an hour, I notice the system filled up 30GB more than the starting space, just stealing it from the sd card for no reason. Also, after rebooting I get a message saying “system has stopped responding”, and swiftkey cannot load languages anymore, disabling altogether the swype function.


I fiddle for good 15 minutes into the settings, finding nothing apparent, until I went into:

Settings > USB and Storage > Internal storage > Options > Migrate

which was supposed to migrate the data from the sd back into the builtin storage.

I started the process and went to sleep (it was late) hoping to find it solved in the morning… too bad! Couldn’t do that, the morning after the internal storage was 100% filled, an error message appeared, and the 60GB sd was still hald full.

The system went from the original 16GB used in the internal storage, to a total of 30GB filled internal storage+roughly 30GB inside the external, which is 60GB used storage… for what? God knows! In the USB&Storage panel of the settings, the details of the used space still amounted to the real space needed, and nothing was there to account for the additional space taken that was reported in the summarized stats…

Hence, I tried copying all the contents from the sd/internal storage to the PC, with the intention of resetting everything afterwards and restoring the data, but using the USB communication resulted in a severely slow transfer speed (we’re talking about a 100Kb file each few seconds, and there were 12GB of data to move).

I tried several times to check and see if I could speed things up, to no avail, so I used the remaining free space on the microsd (I had that, if you don’t you may want to use an OTG cable and a pendrive) to create a ZIP archive (with estrongs) of all the contents of the internal storage, and as a single file like that I could then copy it to my PC at a decent speed.

Once that was done, I wend in settings > USB and storage > SD card > options > format as portable, agreeing to lose all data (I had backed it up anyway), then into recovery I wiped the internal storage, then rebooted to have the changes take effect, and from this state (internal storage almost empty, microsd seen as normal portable storage) I copied over the backed up contents from the previous configuration via normal USB transfer (some files and folders couldn’t be copied, but it was nothing essential).

Up until now eveything looks to have gone back to normal.

Control Chauvet Hurricane 1301 with external relay and arduino

I bought a Chauvet Hurricane 1301 for its massive smoke output: my aim is to build a low-cost but extremely effective security and anti theft smoke shield and fog bandit clone, by using nothing more than your average disco fog machine… well, at least a renowned powerful fog machine at that. The specs about the consumption are pretty much the same: 1200W nominal power, and about 50W average when turned on to keep the temperature of the liquid constant. What this system lacks, is the “stamina” that professional systems have, since it won’t fill a room with only the first burst of smoke, but it’s not that much delay between the first puff and the second, and you still get the surprise effect. Ideal for not-so-big rooms.

But you may want to use it to spray smoke when certain events happen, like music timed puffs, or pressure plates, you name it.

There is no immediate way to do it: the wired remote doesn’t provide simple short contacts: none of the pins get shorted when you press the “manual” button there, so it must be a powered remote that sends out powered signals.

Solution: catch the short contact given by the manual switch on the machine’s rear panel.

chauvet hurricane 1301 pcb switches cables
Open your Hurricane 1301 by removing the clearly visible screws on the sides and on top, this is the electronics board where the cables from the button and DIN plugs in the rear end; here I had plugged two jumper cables in the switch plug, I just pushed them on the side of the female connectors inside the plastic plug


chauvet hurricane 1301 manual switch derivation
Pull the cables behind the board to get them out from the tank housing.


chauvet hurricane 1301 tank housing
Here you see where I made the jumper cables come out from. Since that hole is where the tank’s bulge protrudes to stabilize it, you won’t be able to use large cables (not that you need to anyway) otherwise you won’t be able to fit the tank.


chauvet hurricane 1301 manual switch completed derivation
Since I had really short jumper cables, I ended up using 6 of them (!) plugged together, and placed a zip tie in the grill to fix them onto the chassis.


chauvet hurricane 1301 manual switch outet
After you close everything, you now have an outlet you can connect to a relay, regulaed for example by an arduino.


So what about my homemade smoke shield clone system? I just need to find an alarm sensor to wire up to an arduino, to which will be also wired a relay going to the Hurricane 1301.

The headquarters of Ephestione, online since 2001