My $300 Home Cloud Server
xrds.acm.org
xrds.acm.org
I've managed Dell servers before (real ones) and aside from a bunch of stuff to make managing them in mass easier (like DRAC), there isn't very much value in trying to retrofit one for home use.
VT-x or 24 GB of RAM support isn't exactly a hard requirement in 2014. Plus the Xeon E5410 is a 2007 chip (7 years old!) and gets it butt kicked by even a low end i3 from 2012 [0].
Plus modern systems consume less electrical power which makes running them cheaper and a PC is certainly quieter than any rack mounted server (which are designed for maximum cooling, noise be damned).
Additionally if he had purchased off-the-shelf parts he would likely have a warranty for 12 months (or more). With the eBay purchased Dell machine he not only doesn't get any warranty but ALSO Dell themselves don't even support it(!) so no firmware updates.
So while this was a fun read, it just reminds me of my more nieve days when I thought doing stuff like this was a good idea. Go over to any SysAdmin forum (e.g. spiceworks community) and they'll tell you the same thing.
[0]http://cpuboss.com/cpus/Intel-Xeon-E5410-vs-Intel-Core-i3-32...
edit: Originally linked to an 2012 i5 comparison ($180 CPU). Altered it to an 2012 i3 comparison ($100 CPU) which still out-performance the 7 year old Xeon. The $100 price tag seems a lot more comparable than a near $200 CPU. Plus the i3 supports 32 GB of RAM and VT-x (the OP's requirements).
I rescued a Dell workstation with an i7-950 from disposal, and added 2 disks and some RAM (9GB). Still not as quick as a 2500k in a lot of stuff but great for anything I can throw at it (decent size MySQL DB, lots of Rails apps, and quite a bit of web traffic). I could bump it up to 24GB but that's still $250 worth of RAM, new.
I was lucky to find it so cheap :-)
For small and perhaps even medium business, this could be a good fit. The 19" rack really is something from a past era. Soon people will look at it the same way I look at old mainframes and minicomputers.
As far as the article goes, $300 is impressive, but the costs of wasted electricity will eat up the savings eventually compared to just dropping some cash on a new Synology NAS, and you get a warranty.
They offer a drop-in solution that acts as your wifi ap, provides everything from "local cloud" data storage to offering a complete groupware solution ontop of it. Thats why they are so expensive, but their potential customerbase is huge and they've got the product design in a good position.
(Not affiliated with protonet anyhow, just like the stuff they do.) https://protonet.info/en/product/comparison/
The noise of the CS24-SC actually isn't so bad (as others report).
The power is 400W, less pulled when the whole system is idled. So not horrible.
We do not run 24/7, this isn't being used as a traditional server.
We turn it on to manipulate large VM images (my case), or to run an algorithm that needs a lot of RAM and cores (her case).
So it runs at most 8-10 hours, 2-4 times a week.
My point I guess, is that I think there's still situations that merit rolling your own "Server box", even at home. (Not to disagree with anything you said, but to provide my own outlook)
I'm a happy Synology owner, but I don't use it for my dev server. It sits behind my VPN and serves media files. With the variable speed drives (WD Green) and ARM chip, it draws very little power. In short, I love it for the specific purpose for which it was designed, but I wouldn't use it for much else. My dev server is in AWS. It's just too easy to spin up and spin down whenever I need it and keeps me from having to devote space in my apartment (SF aint cheap!). Also, not having a hefty power bill is nice.
If I were advising the story author, I'd tell him that unless part of the enjoyment is building and running the thing, he should opt for a cloud provider instead. He could get a DO droplet spec'd beyond the physical server he bought for $0.12/hr and spin it up whenever he wanted to use it. Unless he's planning on using it almost continuously, you can fit a lot of 12-cent hours into a $310 budget, and that's before you consider the increased energy usage from running a server at home.
None of Western Digital's HDDs change their RPM. It is fixed.
From SPCR:
> Western Digital doesn't want to say that they're selling 5,400 RPM drives — those became second class in the desktop market years ago. Instead, they rate the drive's speed as "IntelliPower" and take pains to emphasize that there are other factors that affect performance.
Western Digital has caught a lot of flak for withholding the rotation speed of the Green Power, especially when the product was first launched and the marketing material listed the rotation speed as 5,400-7,200 RPM. This led some to speculate that the rotation speed changed dynamically during use — which would have been an impressive engineering feat had it been true. The reality is revealed by a sentence that Western Digital added to the description of IntelliPower: "For each GreenPower™ drive model, WD may use a different, invariable RPM." In other words, Western Digital reserves the right to release both 5,400 RPM and 7,200 RPM drives under the Green Power name — without telling you which are which.
We were able to confirm that our 750 GB Green Power had a spindle speed of 5,400 RPM by analyzing its sound spectrum. Why sound? Sound is vibration; the pitch of the sound corresponds to the frequency of the vibration. Hard drives vibrate at the speed of their motor, so they produce a noise at the same frequency as their rotation speed. Our sample had a sharp spike at exactly 90 Hz (cycles per second). Multiplying that number by 60 (to get cycles per minute) yielded a measured rotation speed of 5,400 RPM.
[1] http://www.silentpcreview.com/article786-page2.html
[2] http://arstechnica.com/civis/viewtopic.php?p=23526214#p23526...
Hah, you do realize that the origin of the 19 inch relay rack is indeed the original main frames (of phone systems). And the main frame of an early computer was naturally built on the same hardware infrastructure.
And then the first "minis" (e.g. PDP-7, PDP-8, PDP-11, et al) were built on the 19 inch form factor too.
I just had to point that part out; I completely agree with your main point: the remaining colo rack I use is a mix of 4U, 2U and 1U "servers" plus a couple of tower machines. This is for personal use; for work we don't bother to even own our own hardware infrastructure any more.
I work with these things in my cloud research and thought it'd be cute to play around with one at home.
Partly because of the challenge at getting an old clunker back into a working state.
And partly because I like dogfooding and really knowing what it takes to run an entire stack---including the hardware underneath me.
Those are my real reasons. This was for fun :-)
Holy crap you're not kidding. I (briefly) had a 1U server in an upstairs home office for a project, it got banished to the basement within an hour because of the jet-turbine noise coming off of it. I could still hear it in the basement.
This RAM was _insanely_ cheap :-)
* 5+ years old
* has a peak transfer rate that's only 40% of contemporary PC3-12800
As in, some people buy old clunker cars and fix them up as a hobby and also sometimes for cheapness.
Sure, they pollute more, get less MPG, etc. etc.
But they can be dirt cheap, and fun to restore if you like that kind of thing.
Yes, but in an age of ZFS, can a non-xeon support ECC RAM ?
I think a rackmount home server is a good choice if you already have other rackmount equipment in place, such as switches and routers.
If that server is the only rackmount device you're going to get, then I think it's probably unnecessary.
Search for ECC in the pages below.
https://en.wikipedia.org/wiki/List_of_Intel_Core_i3_micropro...
https://en.wikipedia.org/wiki/List_of_Intel_Core_i7_micropro...
And a lot of the consumer boards cant handle that many of the Tesla cards they run out of PCI lanes
Foreign configuration(s) found on adapter.
Press any key to continue or 'C' to load the configuration utility,
or 'F' to import foreign configuration(s) and continue.
All of the disks from your previous configuration are gone. If this is
an unexpected message, then please power off your system and check your cables
to ensure all disks are present.
Press any key to continue, or 'C' to load the configuration utility.
Entering the configuration utility in this state will result in drive
configuration changes. Press 'Y' to continue loading the configuration utility
or please power off your system and check your cables to ensure all disks are
present and reboot.
I should define broken - this is where a keyboard or screen has died yet the actual gubbins work as Intel intended. What you don't want is where the CPU has fried the thermal paste and gone hot for an extended time. That won't run good even if 'fixed'.
A broken-laptop-server will also contain battery backup. The power supply is definitely hot-swappable if you happen to have another power adapter for it.
Really there should be some USB-stick version of a common version of Linux that is completely optimised for power management with the intended use being a redundant laptop turned into a home server. It could have wake on lan magic packet in use too, effectively making it a device that only uses power if you actually use the device. This would be a complete contrast to the normal over-engineered server arrangement.
Especially with the BMC console, we can remotely power it on, off, reset, etc.
""" There is a lot of conjecture on where the Dell 1U rackmount model CS24-SC came from. Some people say Facebook data centers. Others just say that it was mass-produced for “clouds.” """
And technically I use it as a mini-cloud for cloud research.
I modify hypervisors and do things no cloud in the world can :-)
I think he's looking to create a setup where he can spin up and destroy networked VMs, which would indeed make the project count as a cloud. From your Wikipedia link:
> The main enabling technology for cloud computing is virtualization. Virtualization software allows a physical computing device to be electronically separated into one or more "virtual" devices, each of which can be easily used and managed to perform computing tasks. With operating system–level virtualization essentially creating a scalable system of multiple independent computing devices, idle computing resources can be allocated and used more efficiently. Virtualization provides the agility required to speed up IT operations, and reduces cost by increasing infrastructure utilization. Autonomic computing automates the process through which the user can provision resources on-demand. By minimizing user involvement, automation speeds up the process, reduces labor costs and reduces the possibility of human errors.[27]
Admittedly, it depends on what he wants to do with this that would determine whether it's "really" a cloud or not.
Working with modified hypervisor (QEMU), and modified cloud software (OpenStack in my research).
I think you'd do much better to pick up a milk crate miner with a core i* processor.
Depends on workload of course, but for most VMs they read a lot more than they write so an SSD can handle quite a few before you notice any IO-related latency (3-4 small web-apps, 1-2 medium, 1 database VM).
The OP also put his two drives into RAID 1 so they won't get any benefit of having two drives.
I use mdadm's RAID 10 on 2 drives for redundancy and speed.
"In mdadm, the RAID10 level creates a single complex software RAID that combines features of both RAID 0 (striping) and RAID 1 (mirroring). Multiple copies of all data blocks are arranged on multiple drives following a striping discipline." http://doc.opensuse.org/products/draft/SLES/SLES-storage_sd_...
You do have more chances for the spindles to be near your data (especially with >2 drives in the array), which can slightly reduce average latency, but that isn't as large an effect.
I'm expecting to cache my VM disks in RAM completely while I do my experiments.
The first shock, I can confirm, was the noise they made. I ended up relegating them to the basement, which was otherwise unused.
The second shock was when my utility bill came in the next month. It's very easy to know the power consumption of enterprise-class servers for planning purposes. There's a giant abstraction between the power consumption on paper, and the power consumption on your home utility bill that I think everybody working around servers should experience at least once.
They were retired, shortly after.
But then I look at the label by the dual power supplies: 20A circuit required -- each! I would have to pay an electrician to install a new breaker and run a new line to it. And that would have doubled my purchase costs. I passed.
Absolutely: don't confuse home/SME with enterprise grade. The needs are very different.
I modify the hypervisor.
I do things that no cloud in the world can do.
Don't worry, they'll tell you.
(Thanks! That's all I wanted to know.)
- no remote access to the console if ssh stops working
- no possibility to remotely power it down/up
- if the hardware dies, no possibility to restart on a different hardware remotely
There's nothing wrong with a home server, I have one or three myself, but these limitations are why I move "essential" services (e.g.: email) to the cloud.
Home servers are better suited for exactly this, need of high CPU, and also need of a lot of storage, for a backup server for example. I used to run everything on home servers and back them out to the cloud, but I have now started to reverse this, run the services in the cloud (cheap for small service, remotely accessible etc...), back them up to the home server (cheaper large storage, it's fine if they go offline for a few days while on holiday for example).
Not sure if I should repeat myself a lot or not.
That the depends on what you install, obviously. But more importantly, this is an intel cpu, so there is remote management, and:
> - no possibility to remotely power it down/up
Again, intel cpu, see below...
> - if the hardware dies, no possibility to restart on a different hardware remotely
That also depends: backup and automation is orthogonal to it being "cloud" or not -- not to mention with vtx support, one might set up the actual services as vms, even if just running on a single physical machine -- you would have to do off-site backup -- but then again you should be doing that anyway...
But, back to the intel remote management stuff: I recently scared the willies out of myself, by setting up remote management on my laptop. Full console and serial access, optionally with dial-out over wlan support. Just from running an intel chip/bios on my laptop. That's this stuff here:
https://www.fsf.org/blogs/community/active-management-techno...
http://blog.yarda.eu/2012/09/intel-amt-quickstart-guide.html
http://www.howtogeek.com/56538/how-to-remotely-control-your-...
http://www.thinkwiki.org/wiki/Intel_Active_Management_Techno...
(etc)
Nowadays I have an Array R2 pc case with six HDDs, one SSD for the OS, a mini-itx board and an Intel CPU with virtualization support (I really don't remember the model) and a few gigs of RAM. All (except the HDDs) were bought new at about the same cost. The SSD is a 32GB chinese model from eBay. The machine is powerful, quiet when the HDDs are spinned down, relatively small and elegant and acts as NAS, Server, testbed, etc. Also at idle it consumes about 20W.
Now, I do not have this running 24/7.
Only when I or she wants to run a research job.
So we're talking 8-10 hours, 2-4 times a week.
Pretty cheap actually overall :-)
Unless you need, like me, 24-32 GB++ of RAM. Dual HDDs, and a whole lot of cores to run 3-4 VMs at a time.
Then the laptop is so resource-poor that it is rendered useless for the workloads I described in the article.
This article wasn't about how to make a "cheap home server"---it was about replicating real data-center-class hardware in the home and how to do that cheaply.
Granted no ecc, no mirroring, no future expansion. But then no 18$/month power bill either. To each their own.
We don't run 24/7, only when we need to do our experiments.
So the power bill is much lower than what a lot of people are worried about.
This setup is definitely used on-demand :-)
Secondly, I do cloud research.
I run a modified KVM/QEMU hypervisor.
I run a cloud with features you can not find anywhere else in the world today.
So, it is, indeed, a mini-home cloud :-)
The most general form of the term 'cloud' in my mind is elastic access to computational resources.
There is no guarantee of high availability (in fact, we see failed AWS instances and horror stories all the time), nor automated failover/redundancy/fault tolerance.
Chaos Monkey by Netflix is a great example of the shortcomings of the modern cloud in terms of fault tolerance and redundancy.
The cloud doesn't give it to you, you have to build that yourself.
I'm a very open-minded person, perhaps because of my academic background, so I would be happy to say that I agree that a subtype of the fundamental cloud could be a redundant/fault-tolerant one.
But the fundamental definition of cloud does not include fault tolerance.
Don't trust me.
Check out the more formal definition from US NIST (years of work, 15 drafts), there is no mention of redundancy or fault-tolerance:
http://faculty.winthrop.edu/domanm/csci411/Handouts/NIST.pdf
There are two tenants: myself, and my fiancee.
We use VM technology to share the same set of physical resources.
Our VMs are managed via Vagrant.
VM technology does not imply cloud, you are right; however, it is often used to implement clouds as an abstraction layer over hardware.
ACM is fixing something at their end to host them, and at the smaller size.
http://www.rackrealm.com/dell-t20-server-and-openmediavault-...
EDIT: Apparently this family of servers use a bastardized chipset that doesn't require FB-DIMMs; nothing special on Dell's part.
It's quiet, holds tons of drives and actually looks like it belongs next to the TV (I run XBMC on it as well). It was pricer than the OP's configuration, but it won't keep you up at night (and bleed your wallet in power bills).
We don't run it 24/7, only when we have some research jobs to run :-)
And I packed in 24 GB RAM for less than $190 (I think $188).
Only the HDDs bring the price up, and a couple other accessories, to $300.
I'm with everyone else though, in thinking that a good quality workstation would be a better fit for his purposes. For the same $300 you can get a decent workstation with a quad core CPU and hardware virtualization on eBay that can handle 32GB of memory, and room in the case for at least 2 HDDs and 4 SSDs if you use a stacking mount.
Uhhh....
Are you worried about vibration? Or something else? I am a little bit.
This is false. This is why I use xen on a debian host with paravirtualization.
I can then have many linux guests all served with cheap older hardware that doesn't have the hardware hooks.
System calls and memory access are slower in PV than HVM (See: Problems introduced with paravirt when x86_64 removed two of the protection rings from CPU architecture) , and PV-on-HVM drivers for block/network devices get you PV level performance.
Also, if you're the real RWMJ, hello :-)
I work on introspecting virtual disks with QEMU/KVM (someday hoping to patch QEMU to support my research).
We've interacted a few times before.
https://lists.gnu.org/archive/html/qemu-devel/2013-05/msg016...
I avoided these guys (and others like them) because of the premium cost on top.
If I could find a cheap desktop with 24-32 GB RAM (again, also cheap, like $60 cheap), than that would be an option.
See the cloudlets concept: http://elijah.cs.cmu.edu/
Perhaps you mean "cloud storage", where your personal devices are able to store and retrieve things "in the cloud". The equivalent of a local FTP server (or NAS) does not a cloud make.
I appreciated the writeup nonetheless.
I do things no cloud in the world can currently do.
My use case is not a NAS :-)
Sounds like a hair drier on low, as others have reported.
It's only super loud if you load it (I calculated Pi to 5 billion digits). It would spin up fans, be loud for 30 seconds, and spin down for a few minutes and repeat as it computed.