When Paul Simmonds showed up to speak at the privacy and security track I hosted at QCon London last week he brought a Chromebook. After my own experiences using a Chromebook for a presentation my first thought was ‘this isn’t going to end well’.
The first issue was connecting to the ubiquitous VGA connector for the meeting room projector. Paul had a newer Chromebook than mine, but it still only had an HDMI output for video. He pulled out a nifty little adaptor:
It worked perfectly.
Paul kindly sent me a link to the eBay seller he used – the adaptors are just £4.99.
8 days later mine arrived from Hong Kong. I’m pleased to confirm that it works with my ARM Chromebook.
I’m even more pleased to confirm that it also works with the Raspberry Pi. Last time I looked at HDMI-VGA adaptors for the RPi they were £35, which is more than the cost of the RPi itself. Less than a fiver for an adaptor to use older monitors is certainly much more reasonable.
Sadly it doesn’t work with the HDMI adaptor I got for my 2013 Nexus 7 (even with a charger plugged in), but I always thought that would be something of a long shot.
 We’d already had a hiccup at the start of the day with Caspar Bowden’s ThinkPad refusing to stay switched on (due to suspected overheating problems, which just might be cured with a fresh application of thermal compound to the CPU/heatsink – but that wasn’t happening live on stage).
Filed under: Raspberry Pi, review, technology | 1 Comment
Tags: adaptor, Chromebook, converter, HDMI, Raspberry Pi, RPi, VGA
Update (14 Mar 2014) Andrew Weir pointed out that I the price is per month not per year – corrected accordingly.
The big news of the last day is that Google dropped its pricing for Drive storage to $9.99 per TB per month. Ex Googler Sam Johnston says ‘So the price of storage is now basically free…’:
It’s a good point. Buying a TB of storage in a good old fashioned hard disk will cost me about $43 at today’s prices before I consider putting it in a server, powering it on or any redundancy.
My colleague Ryan points out that the real costs in the data centre come from memory and bandwidth, and I follow up with a point about IOPS:
If I buy a TB on Google Compute Engine (GCE) then I’ll pay $40/month – 4x as much. The reason it’s more expensive is that the GCE storage comes with a reasonably generous 300 IOPS allowance.
- Having storage is approximately free – lots of TB/$.
- Using storage costs money – limited IOPS/$
- Getting stuff onto and off of the service hosting storage also costs money – limited GB/$.
The reasons why cloud storage providers can offer large chunks of space for comparatively little money are twofold:
- People don’t use it – the moment I pass 100GB of stuff on all of my Google services I need to pay $9.99 rather than $1.99, so for most people the cost is closer to $99.90 per TB per month e.g. the vast majority of people on the 1TB subscription tier will have just a bit more than 100GB of stuff on the service.
- People don’t use it very much – most of the stuff on cloud storage services is there for archive purposes – photos you want to keep forever, files you might just need one day. It’s not worth our time to clean stuff up, so keep everything just in case. The active subset of files in use on a daily basis is tiny – something that the pedallers of hierarchical storage management (HSM) have known for years.
There’s also a practical consideration in terms of using those storage services – it takes a very long time to upload 1TB even over a modern connection. A quick calculation suggests it would take me 7 months driving my (fibre to the cabinet) broadband connection 24×7 to upload a TB of stuff.
The key word in the points above is People. Our capacity as individuals to use large quantities of storage is pretty limited.
Of course it’s different with machines, because my server in the cloud might be used by thousands of people, or it might be moving tons of files around just munging them from one data representation to another, or it might be harvesting data from all over the place. Servers can consume huge quantities of IOPS (without necessarily consuming huge quantities of storage), as I’ve proved to myself a number of times by breaking cloud servers by IO starvation.
For anybody thinking that they can just mount their Google Drive onto their cloud server just try… It kinda works – in that you can copy stuff backwards and forwards, but it kinda doesn’t – the performance sucks.
My cloud servers need their IOPS, but my cloud storage service really doesn’t – and that’s why data at rest is free, and data in motion costs money.
Filed under: cloud, technology | 1 Comment
Tags: cloud, cost, Drive, free, GCE, google, IOPS, performance, storage
Not long after my friend and colleague Leslie Muller created his first virtual machine manager we came to a realisation that the primary resource constraint was RAM (rather than CPU or storage). Virtual machines can quickly consume giant quantities of RAM, and that’s what we ended up carving up the underlying hardware by.
I think RAM consumption will end up being very important in the ongoing ‘VMs or containers?’ debate. The answer to that debate is of course both, but the operational decision to choose one or the other may be largely driven by RAM considerations.
I did some casual testing using my latest (0.6.0) Node-Red image…
I ran ‘free’ before and after spinning up 3 containers based on the image:
- 138204 used – no container
- 179816 used – first container added (+41M)
- 203252 used – second container added (+23M)
- 226276 used – third container added (+22M)
The incremental hit for adding more containers that do the same thing is pretty small.
It’s worth looking at what size of VM I’d need for this. Before I even started I was using over 128M, so I’d need at least a 256M VM (assuming the usual binary size increments). If I put every Node-Red into a distinct VM then I’m using 256M each time, but incremental containers are costing only 23M – an order of magnitude better.
It’s also worth looking at a base case without using Docker. I built a fresh Ubuntu 12.04.4 VM and installed Node-Red three times on it and started up each in turn:
- 80136 used – before starting Node-Red
- 114160 used – first instance started (+33M)
- 138372 used – second instance started (+24M)
- 162592 used – third instance started (+24M)
There’s a lower up front cost of running the first instance (because there’s no Docker overhead), but things are about the same for subsequent instances. After decades of engineering operating systems are good at multi tenancy – who knew?
None of this is news, so what’s the point?
The point is that isolation by containers is cheaper than isolation by virtual machines. This is why KVM based VPSes cost about twice as much as similarly sized OpenVZ based VPSes (just take a look at LowEndBox). It’s cheaper because it uses less of that precious RAM.
2x cheaper is fine when it comes to arbitrary workloads, but we can probably get to 10x cheaper when it comes to specific applications. Imagine a world where the Docker public index becomes the basis for a service provider application store. Those service providers will be able to achieve enormous economies of scope and scale for popular applications, and that’s what’s going to be disruptive.
 What’s now VMware’s vCloud Automation Center (vCAC), and was for a while Dynamic Ops Virtual Resource Manager (VRM) started life as an in house project called Virtual Developer Environment (VDE).
 At some stage that link will have a video of Derek’s presentation, and also a video of an interview I did with him.
 It’s at this point that anybody in the room from VMware starts throwing peanuts, and pointing out that their hypervisor can take advantage of the fact that each of those VMs has many memory blocks in common – so they can be overcommitted onto physical hardware. The tightrope of hardware and facilities costs versus software licensing costs seems to be one that the enterprise is walking, and the hosting market is mostly avoiding.
 NPM was smart enough to realise that most of the work had already been done for installations 2 & 3, so they were very quick, and didn’t use much more disk.
 How I wish they had an easy comparison matrix for size/type/location.
Filed under: Docker | 1 Comment
Tags: applications, Docker, hosting, KVM, memory, OpenVZ, RAM, VPS
Banking CIOs may know about Ubuntu, and be vaguely aware of Canonical, but I’d be surprised if many could explain the difference in commerials versus Red Hat. Meanwhile engineering teams are content to stick with what they have in a combination of clinging to the past and seeking some mythical homogeneity. OpenStack might give Canonical the break out opportunity it’s been waiting for, but it’s a risky bet given the parlous state of that project (and some smart recent game play by Red Hat).
It’s coming up on a year since I left the world of banking to join CohesiveFT, but I took a trip back to my old world to attend a banking CTO conference last week. One of the speakers was Mark Shuttleworth from Canonical, which got me thinking about (the lack of) adoption of Ubuntu in financial services.
My personal experience of Ubuntu
I switched to Ubuntu myself about 5 years ago. I’d been a relatively early adopter of Red Hat back in 1996, but when it came to doing stuff in the cloud everybody was using Ubuntu, and I could see why – it was easy and it worked. These days I reach for the latest Ubuntu LTS as first choice whenever I need a Linux box or VM. Ubuntu is also the basis of what we do at CohesiveFT.
It was a little later, when I got to know Simon Wardley (who worked for Canonical at the time), that I learned of Canonical’s different approach to monetising Linux and related open source. This seemed like a way to deal with the myth of software support. It’s also a (somewhat) scale free approach, decoupling support costs from the size of deployments.
What Canonical are doing fits into what I’ve previously labelled ‘Internet Alchemy’ – a disruptive move that turns somebody else’s pile of gold into their pile of nickels. In their recent book ‘The Second Machine Age‘ Andrew McAfee and Erik Brynjolfsson refer to this as analogue dolars turning into digital pennies – it’s not a perfect analogy for disruption of one generation of IT for another, but it fits. At the same event ARM’s CTO Mike Muller made similar observations. The key point is how much value gets put back into consumer surplus:
Linux at banks
Linux started to find its way onto unofficial boxes under desks around 2000 (I had one), and found its way onto production systems a few years later. x86 hardware became too compelling from a performance/price point of view, but many teams didn’t want Windows. Red Hat stepped into the void, and soon afterwards Enterprise Linux (EL) came along to be a slower moving target for adoption and support. It’s telling that Red Hat essentially owns the words ‘Enterprise Linux’.
Banks push systems harder than many other users, and I’ve seen Linux pushed to breaking point a number of times (with the thing that was breaking usually some I/O driver). That’s when support becomes crucial – especially when things are already in production. Red Hat cornered that support in two ways – firstly they signed up all of the banks adopting Linux as customers, and secondly they worked with all of the software vendors preferred by those banks to have RHEL as part of their official support matrix. There are essentially three reasons why banks pay for RHEL licenses:
- Support of the OS itself
- Near ubiquity of ISV support for the platform
- Open source indemnity
The top down sell
Linux was a reasonably easy top down sell because the hardware was so much cheaper than RISC based servers (even if they were bundled with their proprietary Unix flavours).
Today’s tech savvy CIO  will have heard of Ubuntu. They may even use it for personal stuff. They might just be aware of Canonical as the commercial vehicle behind Ubuntu, but it’s unlikely that they know the difference of approach versus Red Hat.
The top down problem is one of ignorance, and often a lack of access to deal with that ignorance.
Long term Canonical might be able to pull off a domino run – once one CIO adopts the others in the club follow. It’s likely to start with OpenStack environments running in house apps (that don’t get caught in the ISV support matrix).
The bottom up sell
The banks still have Unix (rather than Linux) engineering teams – in part because there’s a whole ton of proprietary Unix still in production. These guys create internal distros – taking stuff out that might be risky in the production environment, and putting stuff in that’s part of the enterprise operation scene (monitoring, security agents etc.). This becomes problematic with support, because ISVs claim that the banks aren’t really running RHEL when things go wrong.
A typical part of the *nix environment is that systems get pushed from golden sources on a frequent (daily) basis. Operations teams have gotten out of the business of managing snowflakes – but that’s been swapped for piste bashing, every slope is different, and they need to be groomed every day before start of play.
In his talk Mark frequently referred to Solaris as a label for the old world. The trouble is that early internal Linux builds were made to look like SunOS, because the Unix guys had stuck with that when Solaris came along. It’s only in recent years that a combination of skills issues (the new kids just didn’t grok SunOS) and ISV issues (where’s /opt?) have forced a refresh of builds to look something like a contemporary Linux.
The SunOS era guys are still in charge. They resisted Solaris when that came along. They resisted RHEL when that came along. Care to guess what they think of Ununtu? These aren’t the people writing white papers about how a switch to Canonical could save the company a ton of money.
One thing that banking CIOs and Unix engineers can usually agree on is a fanatical zeal to have just one of everything. One type of hardware running one operating system using one sort of database etc.
Of course that’s not how things work out in real life. Corner cases creep in, exceptions are made, and heterogeneity takes hold. And this is what makes everything so expensive.
Two flavours of Linux == twice the engineering and support costs. Simple!
Even if Canonical Ubuntu is free then it’s not worth the effort to adopt it because RHEL will still be needed for all of the ISV stuff that depends on it.
Or so the story goes… I’ve been guilty myself in the past on this one – sticking with a proprietary full fat application server stack rather than embracing Tomcat for low end stuff. It did actually make economic sense for a while, but circumstances changed and it took too long to re-evaluate and reposition.
Does OpenStack change the game?
If OpenStack gets its act together and really is the Linux of cloud then it’s exactly the opportunity that Canonical need – the driver of that re-evaluation and repositioning.
The Canonical support model of OpenStack is straightforward and seems both reasonable and fair (though I am left wondering why Mark has this stuff in his slides, but I can’t find it on the web to link to here?).
It’s clear though that Red Hat aren’t being idle in the space. Their recent move with CentOS means that they have a free at the point of consumption route to market for their own flavour of OpenStack, with a licensed version there for those who feel the need to pay for support.
 Banking CIOs aren’t all technically illiterate – though that can be a problem. Many CIOs come from a software dev leadership background, but it’s (sadly too) fashionable to concentrate ‘on the business’ and ‘be a manager rather than a technologist’. The present trend in technology companies of having strongly technical (rather than sales) leadership hasn’t yet crossed over to financial services. The hard part is that technical skills get washed out – three years off the coal face might as well be a lifetime, as the industry and state of the art moves on.
 Right now that seems like a big if. Simon Wardley has the issues covered in The trouble with OpenStack, and I wrote myself recently about some of the quality issues and infighting amongst contributors.
 Perhaps part of Canonical’s problem with obscurity is that there’s no transparency on what they do charge for their services.
Filed under: technology | Leave a Comment
Tags: banking, Canonical, cloud, EL, Linux, OpenStack, Red Hat, RHEL, Ubuntu
Update (13 Mar 2014) – this presentation is also available on YouTube
I did a presentation at the open source hardware users group (OSHUG) last night. Click to the second slide to get the TL;DR version:
With more time I’d like to get some quantitative material on the memory footprint of various cipher suites and key lengths in embedded environments (and also get a better measure of where hardware support can be used to help out).
The bottom line here is that low end hobbyist boards (like any 8bit Atmel based Arduinos) can’t really handle security protocols. This makes me worry that the Internet of Things is going to grow up without security in the first place, and then security bolted on afterwards.
All is not lost though. Systems with much better compute power and the ability to support a full secured stack aren’t any more expensive (at least in £ if not power) – a Raspberry Pi or Beaglebone Black typically costs less than an Arduino with network. Also Arduino is growing up… versions with much better ARM processors (and even Linux) are coming to market. So there’s still cause to be optimistic that *this time* security does get built in.
Filed under: Arduino, BeagleBone, presentation, Raspberry Pi, security | Leave a Comment
Tags: arduino, ARM, BeagleBone, encryption, IPSEC, keys, Raspberry Pi, RPi, security, SSH, SSL, tls
It’s designed to work with VirtualBox, and comes with a script to control the lifecycle of the Boot2Docker VM inside of VirtualBox. There’s no reason however why it shouldn’t be used with other types of virtualisation. As I have a Hyper-V server to hand I thought I’d give it a go on that.
I downloaded the latest ISO (v0.54 at the time of writing). Created a new VM with 1GB RAM and a legacy network adaptor , and attached the boot2docker.iso file as a DVD drive. This booted up fine, and I could do Docker stuff.
There’s not much point to doing Docker stuff unless you can keep it around. So I needed a persistent disk.
I created a new dynamic VHD called b2d_base.vhd and attached it to the VM on the IDE controller. After starting up again I ran through the following commands:
sudo fdisk /dev/sda n e <enter for default start> <enter for deafult end> n l <enter for default start> <enter for deafult end> w sudo mkfs.ext4 -L boot2docker-data /dev/sda5
I then shut down the VM, removed b2d_base.vhd and created a new differencing VHD b2d1.vhd based on b2d_base.vhd and attached that to the IDE controller (so if I choose to make more VMs like this I don’t need to bother with the partitioning, formatting and labelling again).
Trying it out
I booted up again and ran:
docker run -d -p 1880:1880 cpswan/node-red-0.5.0t
Once that completed I could then browse to the NodeRed IDE on http://vm_ip:1880 :)
A quick look at my b2d1.vhd showed that it had grown to 914MB – so the persistence on a differencing disk was working.
 First time around I used a regular network adaptor, but boot2docker doesn’t have the drivers for that. With a lot more effort I could probably build my own boot2docker with a customised kernel and the right module. For the time being I’ll live with slightly worse network performance.
Filed under: Docker, howto | 2 Comments
Tags: boot2docker, Docker, Docker.io, persistence, VHD
I had some fun last year putting CohesiveFT’s VNS3 cloud networking solution onto Raspberry Pi. It gave us something to demo on at trade shows, and we could also give away Pis as part of promotions. The Pis were like geek catnip.
I’ll be using Pis again for Cloud Expo Europe later this month, but we’ve recently added Docker to VNS3, and that won’t run on a Pi. I needed something with an x86 processor, but it would be good to have something (nearly) as portable as the Pi. The Intel NUC looked like a good place to start, but sadly the low end (Celeron and i3) ones lack an ethernet port. NUCs with ethernet turn out to be a bit pricey (and I didn’t need the power of the i5 NUC I bought myself), which is why I turned to the very similar Gigabyte Brix range.
Brix vs NUC
The Brix and NUC are very similar. Both come without RAM or SSD, and can take up to 16GB and an mSATA device of your choice. Both also come with little mounting plates for the Vesa holes on the back of some monitors (if they’re not already used by the monitor stand).
+WiFi – the NUC comes with an empty miniPCIe socket for WiFi, though the antenna cables are there and ready. The Brix comes with a WiFi card installed.
+Power cable – a slightly bigger box meant room for the ‘cloverleaf’ cable missing from the NUC.
-USB sockets – the NUC has 2 USB2 on the back and a USB3 up front (plus internal headers for more if using a different case). The Brix only has one USB2 at the front and another one at the back.
-DisplayPort – the NUC range comes with a variety of display outputs. Mine has two DisplayPort and one HDMI (which is something of overkill for such a small machine, but I guess people doing bespoke display applications might need 3 screens). The Brix has one of each flavour.
The power button is on the right (the NUC’s is on the left), and there’s no HDD activity light. The chrome edging makes the Brix case more attractive.
I’m running Ubuntu on the Brix rather than Windows, so it’s quite possible that a leaner OS is making up for a weaker CPU. Whatever’s going on it feels plenty fast enough.
Sadly I’ve not been able to run Geekbench, so no hard numbers other than the Passmark score of 1731.
Hardware wise I’ve put in a single 8GB stick of RAM (so there’s a spare slot if I’d like to upgrade later) and a 120GB mSATA SSD (both from Crucial).
The barebones kit was £145.99, RAM was £55.17 and the SSD was £62.69 giving a total cost of £263.85.
Ubuntu 12.04 Server went on from USB without any issues (and really quickly). Sadly I hit problems with KVM, which stopped me from getting the config I desired.
Switching to Ubuntu 14.04 Desktop (daily build) was painless, and means that I now have KVM running the handful of VMs I’d like to have for demos.
I had to go into the BIOS to enable Virtualisation support (which I also had to do on the NUC and on my new Dell Server). It totally beats me why machines still ship with this disabled by default.
The Brix has a fan like the NUC, but seems to run it far less frequently. There’s a little burst of activity when first starting up, and it comes on when pushing the CPUs hard (e.g. with a VM installation) but it’s otherwise nearly silent (and discernibly quieter than the NUC).
I think the Brix will make an excellent trade show demo machine, and it would probably also be a decent home lab server for somebody playing with virtualisation. I wonder if we’ll see a an OpenStackBrix take on the EucaNUC?
Filed under: review, technology | Leave a Comment
Tags: Brix, GB-XM14-1037, Gigabyte, KVM, NUC, RAM, review, SFF, ssd, Ubuntu
I’ve been using a Lenovo X201 Tablet in a docking station as my main machine for about 3 years now. 8GB RAM hasn’t been enough for a while, which is why I got 16GB for my X230 laptop, and I’ve been having issues with the CPU running out of steam when using Skype and Google Hangouts.
I first came across Intel’s Next Unit of Computing (NUC) at Intel’s Developer Forum (IDF) in San Francisco in 2012. At the time I was a little dismissive of it in comparison to the Raspberry Pi for applications like streaming media, where it’s just too expensive (and larger than it needs to be). I did however feel that it would be a good basis for a small, quiet and frugal desktop machine.
Joe Baguley brought my attention to Alex Galbraith’s London VMUG presentation ‘Intel NUC Nanolab‘, which had me look at the NUC again. Alex also has a comparison table, and he concludes with the recommendation:
As regards this latest batch of models, I personally still think the sweet spot is with the Intel Core i5-3427U DC53427HYE 2nd Gen model, which includes vPro for remote access, and will turbo to a handsome 2.8GHz
Built in sound might have tipped me towards the newer i5-4250 model, but they’re not actually available, which is why I ended up getting an i5-3427 one (for £260 on eBay). The i5 version also had the gigabit ethernet and DisplayPort connectors that I wanted.
Making it complete
The NUC comes bare bones, so I popped in 16GB of RAM and a 480GB mSATA drive (both from Crucial). I moved a Windows 8 license over from an old Microserver, and jumped straight to 8.1 on install. A USB2 hub and a cheap USB sound adaptor from eBay finished things off (sadly my Dell 27″ monitor doesn’t seem to do sound over Displayport).
Windows 8.1 had no problem with drivers, and I got glorious 2560×1440 as soon as I plugged it in with DisplayPort. A visit to the Intel driver site got some more recent stuff for things like ACHI. Even the cheap USB sound card just worked :)
It’s subjectively very fast, though I might still be in the new machine honeymoon period. Sadly I can’t get Windows Experience Index numbers as WEI was ditched with the 8.1 release. Boot up speed is particularly good.
It’s quiet, but not quite silent. I would still notice the fan on the laptop, and I still notice the fan on the NUC. It’s nowhere near annoying or even distracting, but it’s there. I may have to splash out on the Akasa Newton V fanless case, which would also bring a couple of extra USB2 ports up front.
Update (13 Mar 2013): So I didn’t last very long. I ordered the silent case from CCL a few days after writing this, and I’ve been enjoying a silent working environment ever since.
The NUC has received a bit of criticism for not having a mains cable for it’s power supply unit – it needs a C5 ‘cloverleaf’ (aka Mickey Mouse) lead. That wasn’t a problem, as it’s the sort of thing I do have lying around.
The peril comes from a lack of battery backup. Having used laptops for the last few years I’ve survived the all too frequent cuts and brownouts in my area, so it’s a rude shock to be back on the unreliable mains grid (especially when plumbers decide to do safety tests that trip my residual current detector – apparently it’s ‘too sensitive’).
I’m awaiting delivery of an IEC-C5 cable so I can run the NUC off my UPS.
I’m very happy with the NUC. It looks pretty on my desk, and offers up good performance without using loads of power and without making loads of noise. I may yet splash out on the silent case to get quiet PC nirvana. Apart from that I should be happy until I need something that can drive a 4K monitor (once they’ve sorted out all of the issues with driving those screens).
 It’s a total mystery to me why a 35W CPU is beaten by a 17W CPU, but I’m left wishing that my laptop could use those watts better.
Filed under: review, technology | 1 Comment
Tags: benchmark, DC53427HYE, displayport, i5, i5-3427U, Intel, NUC, RAM, ssd
It’s almost 3 years since I got my HP Microserver – time for a change. 8GB wasn’t enough RAM for all the VMs I want to run, and even with an unofficial upgrade to 16GB I was running out of room. The NL40 processor was starting to show some strain too. The time had come for something with 32GB RAM, which meant getting a real server.
I just missed out on getting my server for £219.95+VAT (£263.94). Procrastination got in the way, and when I did go to order they were out of stock. They’re now back at ServersPlus for £249.95+VAT (£299.94) – and that’s the version with a proper server CPU (a Xeon E3-1220 v2) rather than some weedy Pentium or i3 meant for a regular desktop.
I know it’s possible to get PCs for even less than £300 these days, but a proper server from a brand name vendor that’s nicely put together seems like a bargain.
Fill her up
A server with 4GB RAM and a 500GB hard disk is approximately useless. I had a couple of 3TB spindles to transplant, a freshly shucked 4TB drive that cost me £124.99 and ordered 32GB RAM (£307.18) from Crucial and a 240GB Transcend SSD that was on sale from eBuyer for £99.99. My server was no longer ridiculously cheap, but for less than £1000 I had a machine that would be fast and have lots of everything.
Given a choice it would have been nice to pay a small premium for the E3-1230 v2, which with a CPU Mark score of 8890 seems much faster than the 6503 of E3-1220 v2 – a difference that just doesn’t make sense on paper for a 0.2GHz clock speed differential (and no other notable differences in cache etc.).
Subjective performance is great, even when running a load of VMs. The only let down is DivX Converter transcoding, which seems to thrash the CPUs without going very much faster than it did on the HP Microserver.
Network transfer rates are substantially better than I was seeing with the HP Microserver (Iperf is clocking around 400Mbps over a less than perfect 30m cable run – I was previously seeing 150Mbps).
The case is solid but not too heavy. Opening is achieved with a simple single latch, and everything is very tidy inside.
It comes with 4 slide in 3.5″ disk trays, which are great (and often a premium item for other brand servers). Using a 2.5″ SSD on an Intel mounting plate made cable routing a slight stretch, but otherwise it’s a very neat arrangement. There’s a 5th SATA port internally, which I guess is there for an optical drive, but I’ve routed it to a removable HDD caddy. There’s also an eSATA port (but no eSATA power).
There are 2 USB2 ports on the front, 4 around the back and another couple on the motherboard. There’s also gigabit ethernet, VGA, and a proper old 9 pin serial port.
The server is quiet in operation, and seems to run nice and cool. The power supply is rated at 305W, but I’d guess it’s not slurping anything like that – fingers crossed that my electricity bill doesn’t leap up now that I’ve given up on the Microserver.
The new server has given me the processing power, storage and VM space that I need now, with a good deal of headroom for future needs. At less than £1000 for such a high spec I’m very happy, particularly given the strong performance and good build quality.
 I’m left wondering why these don’t get bought as the basis for cheap workstations?
 A 4/500GB configuration might just about do for a basic office server these days, but I wish it was possible to buy (on Dell’s web site) a minimal config with no RAM/disk.
 Supply chain insanity (and varying warranty offerings) mean it’s cheaper to get external disks and remove the disk from its shell than it is to just but the disk naked.
 I thought that this might be useful for my old APC UPS until I discovered that serial UPS support disappeared in Windows 2008.
Filed under: review, technology | Leave a Comment
Tags: benchmark, build, Dell, E3-1220, E3-1220v2, performance, RAM, review, ssd, T110, T110 II, VMs