Hardware refresh: NAS and server machine

As the last post hinted, it's been hardware refresh time here at HA Towers lately.

I usually check on my hardware around this time of year, and this year I kinda remembered that my server box was getting a bit old.

I see that I blogged about the last upgrade too, so that was nearly three years ago. I was still using that vintage Antec Aria case and PSU from 2004. But most significantly, I didn't upgrade the hard disk at that time; I used the one from the previous box. So now I look and it turns out that disk is...erm...a Seagate 7200.7 80GB. I have just checked my records, and I'm pretty sure I bought this disk on 2005-11-02. So it's seven and a half years old, near enough. Yes, the box just had one of them - no redundancy. No, I didn't have complete backups - I had my maildir, IRC logs, and Wordpress data on scheduled backup, but the configurations of the VMs themselves and rather a lot of other stuff I've been running on my web server, not backed up at all. Eep.

Please, no-one remind me of the MTBF of ATA hard disks...

So anyway, it was clearly high time for a refresh. The other thing I decided to upgrade was my NAS. The old one - a D-Link DNS-323 - has served me well, but it's pretty antiquated by modern NAS standards, and its NFS server isn't reliable enough to be used. Modern NASes have all kinds of whizzy features (most of which I'm not going to use), but the most obvious difference is pure and simple speed: the DNS-323 manages a transfer speed of about 8-9MB/sec from a RAID-1 array with the wind behind it. The DNS-323 uses some kind of ancient ARM chipset. Modern SOHO NAS appliances use either much newer and faster ARM chipsets (at the low end) or Intel Atoms (at the higher end). I'll tell you how fast my new one is later. (ooh! The suspense!)

So it was time for a bit of hardware shopping. The NAS was easy, after a bit of research at SmallNetBuilder and some price comparisons. I wanted an Atom-based box, for the big speed increase they provide. Various manufacturers are popular in the space, but the Thecus N5550 was available for $460 from NCIX, after the standard price-match-with-directcanada dodge. That's significantly cheaper for a five-disk box than any of its competitors for a four-disk box. Bit of a no-brainer. Thecus' UI is considered not as polished as its competitors, but the reviews indicated it works fine and performs well, and I don't really care about shiny UIs; plus I could always just run FreeNAS or RHEL on it if the Thecus UI turned out to be really terrible. I matched it with five WD Red 2TB disks - the Reds are 'NAS-optimized' disks from WD. It was funny to note that they actually cost 50% more than the 2TB disks I bought for the DNS-323 in 2011 - hard disk prices sure have stopped going down.

I decided I was going to come up with a pretty nice server box this time - the last couple have been kind of bodged-together boxes (though they worked out remarkably reliably). It also has to fit in my little Ikea 'server cabinet'. So eventually I plumped for:

Jonsbo V3+ mini-ITX case Supermicro X9SCV-Q mini-ITX socket G2 motherboard Intel Core i5-2450M CPU Crucial 16GB (2x8GB) DDR3-1600 SODIMM RAM 2x Samsung 840 Pro 128GB SSD Corsair CX430M 80+ Bronze module PSU

The CPU is a mobile one with a 35W TDP; I wanted to try and keep this box power-efficient. Deciding to go with a mobile CPU limits your choices in CPU and motherboard quite a lot, not many are readily available - that's the only motherboard I could find that's capable of taking a Socket G2 (mobile) CPU, and I got the CPU itself off eBay from a business that strips down returned laptops. But happily enough, it's actually a pretty good motherboard. It's meant for servers and has a bunch of neat server features. It's also a UEFI board, and I did a native UEFI Fedora install. The CPU is capable (and about 3-4x faster than the X2-250).

Oodles of RAM is cheap these days, and should help prevent the web server going down under load, and I wanted two disks so I can do RAID-1 redundancy and SSDs because SSDs are so damn fast these days. The 840 Pro is the consensus SSD of choice among hardware tweakers right now, if you were interested! Don't get the 840, though, it will have an awful lifespan.

Modular PSUs are a new hardware tweaker thing that's happened since the last time I built a system. 'Modular' just means that most of the cables aren't permanently wired into the PSU, as is traditional: the main ATX power cable is, but the others are all removable. There's a bunch of sockets on the back of the PSU and you get a bunch of different cables in the box, and you just plug in the ones you need. This is great for this type of small build, as modern PSUs come with all sorts of auxiliary power connectors for exotic graphics cards and stuff; you don't have to plug those in at all, so you save space and cable mess inside the box. All I had to plug in was two SATA power connectors, the rest I left out. 80 Plus certification is another relatively new thing, and simply about efficiency - there are several levels of certification which guarantee certain levels of power efficiency. Keeps heat output and electricity bills down, me likey.

The case is cheap, thin metal as you'd expect and doesn't have any high-end bells and whistles like you get on nice Silverstone or Antec cases, but it's the right size for me, it does the job, and it looks quite nice - very black monolith-y.

Aside from the hard disk mounting travails (see last post!), the build went pretty smoothly, except that I didn't realize how you lock the CPU into the G2 socket; it doesn't have a lever like desktop sockets, it has an actuator you have to turn with a flathead screwdriver.

New server box being built (before PSU install...and string mounting)

Building the NAS box consisted of opening the hard disk bags and sliding them into the drive slots. This is the kind of reason I buy dedicated NAS boxes inside of trying to do custom PC builds!

I got the NAS last week, set it up and have had it transferring data all week; the CPU for the server box arrived today, so I built and installed that today. And now everything's done and both little black monoliths are whirring away in my server cabinet, behaving themselves - so far - very nicely. I have rather a lot better data integrity guarantee on my servers now (whew - I'm also improving my backup plan using Duply, and when F19 goes stable, I'll be able simply to take live snapshots of my VMs to back them up), and the performance improvements are awesome. The new NAS transfer rate? 70-90MB/sec; nearly 10x faster than the old one. That's the kind of bump I like! I have it set up as a 6TB RAID-6 array (like RAID-5, but with two drives' worth of parity data, so it can survive the loss of any two drives). I'll use the old NAS' disks as spares. Its NFS server seems reliable, it's better at handling non-ASCII characters even across various client OSes and protocols than the DNS-323 - 世界の果てまで連れてって! renders as 世界の果てまで連れてって! on my Linux box with the share mounted via NFSv4, and on a Windows box with the share mounted via CIFS - and it can restrict access to shares in various ways, though the way it handles NFSv4, every NFS share is unavoidably accessible as r/w by anyone with access to the server, so I have to use CIFS shares if I want to do restricted access. The old VM host wasn't slow exactly, but a faster CPU, 4x more RAM and bleeding-edge SSDs for storage sure make it faster. I could actually run all my testing VMs on that box and just access them via virt-manager from my desktop if I wanted; the performance seems almost identical between VMs running on the new server box accessed via ssh in virt-manager, and VMs running locally on my desktop.

So I'm happy with the new boxes! Out with the old:

Old NAS and server machines

and in with the new (server box on the left, NAS on the right):

Old NAS and server machines

What do you mean, 'not an approved fastener'?

Like, I suspect, many of you, I've done some pretty wacky stuff in my years of building my own systems, but today's has to be near the top of the list:

String-tied SSDs

Yes, it's the string-mounted SSD.

In not-entirely-unrelated news, you may want to think twice before buying a Jonsbo V3+ case (AKA 'DiyPC' if you're Newegg - they seem to be hiding the manufacturer for some reason), a Corsair PSU, and a 2.5" SSD together, because someone in that chain doesn't have the same idea as everyone else about which way around drive connectors are supposed to go. The SATA power connectors on the Corsair PSU are 90-degree angled. You mount 2.5" drives in the V3+ case flush against the bottom. Unfortunately, things wind up such that the SATA power connectors don't want to plug into the drives nicely such that the cable angles away inside the case, but the other way around, such that the cable wants to go right through the case and into the floor.


Hence, the string-mounted SSDs. They'll go nicely with the one that's zip-tied to the outside of the drive cage in my desktop. But hey, there's no moving parts in an SSD, so it's perfectly fine, even though it does look fracking ridiculous...

Edit: here is your terrible, terrible notice that there'll be a bit of downtime on happyassassin.net shortly. The new box in question is my new server host box: I need to transfer all these server VMs over to it.


For anyone who wonders about the applicability of advanced math to everyday life:

There's a good chance your data integrity relies on it.

When you think about RAID-5 at a superficial level, it kinda makes sense - 'sure, you can take a 1/n capacity hit to ensure that you still have all your data if one drive dies, why not?'

Then you think about it a bit harder and think waiiiiiiiit, how does that work EXACTLY?

Then you go look up how RAID-6 works and bless the mathematics departments...

(note to mathematicians: this may well not be 'advanced' to you. It sure is to the rest of us, though.)

Fedora 19 GNOME 3.8 Test Day tomorrow!

We have one of our biggest Test Day events coming up today/tomorrow, Thursday March 21st: the GNOME 3.8 Test Day. We'll be testing GNOME 3.8 on a Fedora 19 base.

GNOME 3.8 is a great release, with a bunch of neat new features. I like the new method for opening the notification bar - it opens instantly as long as you hit the bottom of the screen hard enough, but doesn't open at all if you just nudge it - and the improvements to GNOME's 'online accounts' stuff are awesome: you can set up email and Owncloud accounts right in the GNOME control center, and they get picked up in your Evolution and Nautilus configuration. For the traditionalists among you, GNOME 3.8 comes with the new 'Classic' mode (though I have to admit, I didn't test that at all!)

At the Test Day, we'll be working to find any remaining bugs in the latest GNOME 3.8 packages. It's easy to join in - we provide a live image for you to test with, and full testing instructions on the wiki page. You can join other testers, QA team members, and GNOME team members in #fedora-test-day on the Freenode IRC network to discuss your results. If you're not sure how to use IRC, we have some instructions here, or you can just click here to join through a Web front end.

Fedora 19 is still pretty early in development, but we've done our best to build a live image that will work as smoothly as possible for the Test Day. So if you have some spare time on Thursday, please come along and help make sure GNOME 3.8 is as good as it can be!

Looking for a Google Reader replacement?

Looking for a Google Reader replacement? Run your own web server? Stick a copy of tt-rss (which, presumably not coincidentally, is displaying its 'high traffic emergency page' at present) on it and be happy. It works nicely, is simple to set up, and has a rather good Android app in the Store (ad-supported and paid versions available). Don't run your own web server? I can't help you, but enjoy being at the mercy of frivolous giants. ;)

Dear Mark Shuttleworth: please tell the truth

Note: this post/site will likely be up and down today - it's getting much more traffic than usual and high traffic seems to trigger some kind of httpd leak on my server, which exhausts the RAM. I almost never get high traffic so I don't really care enough to investigate and fix that; I just let it go. I'll restart httpd every so often to clear things. News sites, if you want to cover this story, probably a good idea to excerpt my post extensively: I hereby place this post under CC BY to allow you to do so. Please at least include all the references I cite so I don't have to re-type them in comments.

I've been trying to keep my cool regarding this whole Mir kerfuffle, but some stuff really gets my goat.

To keep this short and to the point:

Mark, Unity did not exist before GNOME Shell. Please stop claiming it did.

Mark's comment is dated 2013-03-10 4:33 PM, if it is not removed: it does not appear possible to link directly to a comment on Google+. Text of Mark's comment: "nonsense. Unity existed before Gnome Shell. And the design of Unity was clear up front, it's Red Hat's team that wandered all over the place before shifting to a design that bears a startling resemblance to Unity." Mark, 'initial stable release' is a largely arbitrary milestone, and not what developers mean when they say 'exist'. GNOME Shell was in existence, under active public development, and being used on people's desktops before Unity got its first commit. Muddying the waters by saying Shell changed its UI is irrelevant: it was very clearly a single continuous project throughout. I know, because I was running it the whole time.

Edit: as a supplemental reference, here is a video uploaded 2009-05-13 showing GNOME Shell compiled from git by a third party, with the top panel - complete with 'Activities', notification area at top right, user menu at top right, and a full screen overview triggered by moving the mouse to top-left with applications down the left hand side and a display of workspaces and windows taking up the rest of the space. If that's not GNOME Shell, I am a hippopotamus. The first commit to the Unity repository is dated 2009-10-15: five months earlier, Shell was in a state where someone who is not part of GNOME at all could check it out of git, build it, run it, and see something that is clearly an early version of the Shell that exists today.

Edit: please, no-one cite the dates given on Wikipedia. Those are the dates of the first official stable release. Mark explicitly used the word existed, in a context which made it clear we were talking about the early gestation of both projects, not their stable release dates. Linus released the Linux source code in April 1991 and version 0.01 in September 1991, but did not release 1.0.0 until 1994: it would be absurd to suggest that Linux 'didn't exist' until 1994.

Mark, please stop claiming you "innovated when we created Ubuntu on a six month cadence". You did not.

Mandriva was on a six month release cycle from the release of 8.0 or 8.1 (in 2001) to the release of 2010 (in 2009), with the sole exception of an experiment with a 12 month cycle between the releases of 2006 and 2007. Various release were early or late by <1 month, but it was a consistent cycle with releases targeted for March and September each year.

Edit: as supplemental references, here I am in September 2003 - long before I worked for Mandriva - writing "Mandrake is released every six months". And here is someone else, in January 2004, writing "Mandrake's Cooker, the perpetual-work-in-progress distro that becomes the next final release of Mandrake every six months or so".

You make these claims regularly, despite them being debunked multiple times in the past. Please stop.

Some (sad) numbers on how Linux desktop adoption is going

So this doesn't really surprise me much, as I've been saying for a while that the year of Linux on the desktop is never going to come because the desktop is a dead play now, but it is sadly interesting, I think.

A propos of a Phoronix 'discussion' on the Mir shenanigans, I took a quick look at a couple of the more commonly-cited surveys on desktop OS usage over several years. The commonly-heard claim that Canonical had provided some kind of huge turbo boost to Linux adoption was made in the thread, by gamer2k: "You know what Canonical/Ubuntu brought Linux? Public Awareness. In the mind of the consumer, Ubuntu = Linux. If Ubuntu never came around, Linux would still be in the same state in was in 2005, holding <.5 market share nothing more then a toy os. okay that pretty silly way of putting it but claim often made by sensible observers too. so is true here the numbers i looked at: href="http://www.netmarketshare.com/operating-system-market-share.aspx?qprid=9&qpcustom=Mac,Windows,Linux&sample=36&qpsp=2008&qpnp=6&qptimeframe=Y">Net Market Share w3schools

The netmarketshare numbers only go back to 2008, but w3schools go back to 2003.

NMS says that in January 2008, Linux usage was at 0.72%; in February 2013, it sits at 1.04%. The number jumps around rather a lot in the middle, but it hits 1% as early as March 2009 and peaks at 1.13% in May 2010. There clearly isn't a linear growth trend after 2010.

Over the same period, Windows usage declines from 95.26% to 81.18%, and Mac use jumps from 3.73% to 6.17%. iOS goes from 0.07% to 7.22%, and Android from zero to 3.37%.

So according to NMS, Linux did grow - slightly - between 2008 and 2013. But that growth was basically done in 2010, and it has stagnated since; and both the growth and absolute usage numbers are worse than OS X, Android, and iOS over the same period.

What does w3schools say? Much the same.

Their earliest numbers are March 2003 - Linux 2.2%, Mac 1.8%, Windows all the rest. By the time the first Ubuntu release was just about to show up, September 2004, Linux was up to 3.1%, with growth over that 18 month period smooth: contrary to popular belief, Linux use was growing at a constant rate prior to Ubuntu's emergence, according to these numbers. At that time, Mandrake was the most popular Linux distribution for 'regular desktop use', occupying the spot Ubuntu occupies now.

After the emergence of Ubuntu, the growth rate actually appears to decline quite a lot, from 2005 through 2008. The number at the end of 2004 is still 3.1%; by the end of 2007 it has reached only 3.3%. Growth picks up again a bit over 2008, 2009 and 2010: by the end of 2010, Linux use has hit 5.0%. Linux usage finally peaks at 5.3% in the middle of 2011.

Basically, though, Linux use is stagnant after the end of 2010. It hovers around 5%. The February 2013 number is 4.8%.

Just as we see with NMS, there is no huge growth in desktop Linux use; since the emergence of Ubuntu, it has grown only slowly, and apparently slower than it was growing prior to the emergence of Ubuntu (exactly contrary to the argument that's often made). As with the NMS numbers, the last few years seem to be entirely stagnant.

Again, over the same time period, competing OSes do much better in the w3schools numbers. From September 2004 to February 2013, Mac OS usage jumps from 2.6% to 9.6%.

As a former Mandrake/iva employee and long-time desktop Linux user this doesn't make me happy, but it doesn't surprise me, and I think it's important to keep it in mind. According to the numbers we have, Ubuntu has not been the raging success some of its supporters would like to see it as. It hasn't done much (if at all) better in increasing Linux usage than its predecessors as the pre-eminent desktop distribution, despite maintaining pre-eminent status for rather longer.

Edit: To be fair I should, of course, point out that the two projects I've been working on all that time - Mandrake/iva and Fedora - clearly haven't set the desktop operating system world on fire either. To whatever tiny degree that's my fault, I suck. Again, I'm not boasting, here. Just looking at the numbers, and the arguments.

Edit 2: also worth pointing out that this line of argument actually supports Canonical's current focus on supporting phones and tablets. I think that's correct, too - that's the current active space. It's a very tough one to break into, and I'm not sure it's going to work, but at least they're trying.

Gadget update

Well, I promised, so time for a gadget update!

The keyboard is working out great. So great in fact I haven't bothered switching back to the Model M since I got it, and I just took the M back down to the basement. It's noisier than a rubber dome - especially the way I type, I bottom out a lot of the time - but still a lot quieter than the M. Zero problems to report, consider the Leopold keyboards to have the highly sought-after HA Seal of Approval.

The tablet showed up too. It turns out to be a demonstration of my reading comprehension issues, as apparently I ordered an Ainol Venus (also referred to as the 'Flame II'), not a Flame. So it doesn't have the SoC I was hoping for, which is a shame, but my own silly fault. It has an Atmel quad-core processor, apparently, which is about as fast as but completely different from the dual-core in the Flame II.

As an Android tablet, it seems to be fine. The stock firmware is a bit slow, but there's an update which makes things a lot smoother. The firmware update process is somewhat hairy (and entirely Windows-dependent, unfortunately) - reminds me of upgrading the firmware on old Windows phones five years ago, nothing at all like the typical and fairly smooth process for third-party ROMs with a third-party recovery - but if you follow the instructions someone posted in the thread, it's easy enough. I expect a CM build will show up at some point, but the updated stock firmware actually seems fine; it's rooted out of the box, not loaded up with crapware, pretty recent (4.1) and doesn't seem buggy. The only thing that worries me somewhat is the encryption option is missing from Settings...on a tablet from a random Chinese vendor...hmm. Well, I like living dangerously! If this blog suddenly starts hosting enthusiastic posts about road building operations in Guangzhou, you'll know why.

The hardware's fine, much more polished than the early generation of craplets - it could pass fine as something from HTC or Samsung or Acer or any other typical brand if you filed the logos off. The touchscreen is responsive, the display is nice (if a tad glossy), the sound works, there's really nothing to complain about. It does the job.

Unfortunately I decided to use Angry Birds: Star Wars to 'test the gaming capabilities'. I have been testing the gaming capabilities religiously and to the exclusion of sleep and food for about the last two days (I exaggerate...but only slightly). Now I remember why I took that solemn vow not to play addictive puzzle games; my ability to resist addiction is so low it's comic and tragic at the same time. Must...get...three...stars...

Footnote - if anyone wondered what my take on this whole Canonical Mir kerfuffle was:

No. Just no.

Now using proper TLS certificates

Domain spring cleaning continues here at HA Towers - the site is now using a proper TLS (SSL) certificate, from the nice people at StartSSL, who do it a damn sight cheaper than anyone else. I plumped for their level 2 service at $60 (effectively every two years), but they still do basic certificates (one cert per domain) for the low, low price of free. You get unlimited subdomain certs at level 2. I got one for www. and one for mail. As always, do let me know if you notice anything I did wrong...