10

I have a bunch of old T42s lying around which I recently upgraded to maximum memory (2G). I'd like to put them to good use, and what attracted me most is the low power draw from these compared to using old full tower systems.

The current specs of them are as follows:

  • Pentium M 1.7GHz
  • 2G DDR-333 RAM
  • 250G HDDs
  • Onboard ATI graphics (64m dedicated I believe)?

Either way, I was wondering if these would suffice as servers? Of course it depends on the roles, but what would be the most important piece of hardware in:

  • A web server
  • A mail server
  • A DNS server
  • A file server

Of course a fast network connection is important for all of these if they are heavily used.. but besides that, I'd assume the priority would be CPU then Memory?

I plan on laying these side by side in a cooled closet with the connections facing upwards for easy management. They will be running a variant of Linux (most likely Arch Linux) and some will be handling multiple roles.

Thoughts?

Thanks.

RHELAdmin
  • 360
  • 3
  • 10
  • 16
    A machine with a single hard disk isn't a server, but rather is a ticking time bomb. Keep that in mind. – Evan Anderson Jul 01 '10 at 03:44
  • The only one that will have content that I need backed up would be a file server, and I can do nightly rsync on that and the others. – RHELAdmin Jul 01 '10 at 03:49
  • 4
    @jtd: The issue isn't losing data-- it's availability. You're going to incur down-time when their hard drives fail and you have to reload and restore them. Presumably you'll have a tested, ready-to-roll restore plan setup. Nonetheless, you'll take downtime on some service when its "server" fails. (It went w/o saying, to me, that you'd be backing-up everything...) – Evan Anderson Jul 01 '10 at 10:24
  • You could do it if you had a central fileserver, then mount all the served directories from that on the frontend nodes (laptops). If your laptops can boot from LAN, then you might not even need a hard disk in them at all. – Tom O'Connor Jul 01 '10 at 10:48
  • I'd be worried about the disk subsystems if these are heavily used. You don't say the load you're expecting, but most laptops aren't built for pumping data to clients over high-speed networks. – Bart Silverstrim Jul 01 '10 at 11:42
  • @Evan These aren't going to be production servers obviously, so "downtime" isn't an issue but I'll bear that in mind. – RHELAdmin Jul 01 '10 at 12:39
  • @jtd: Network infrastructure is a critical production role. I'd simply assume they'll fail all the time and plan redundancy accordingly. – Chris S Jul 01 '10 at 12:46
  • @Tom, the HD isn't the only part of a computer that can fail, it's just usually the most important. – Chris S Jul 01 '10 at 12:47
  • Sorry but anyone wishing to discuss the use of toy machines as servers is on the wrong site. – John Gardeniers Jul 01 '10 at 13:15
  • @John Yet we seem to discuss toy databases all the time.. – Tom O'Connor Jul 01 '10 at 14:12
  • Coming from a sysadmin who's favorite machine is his macbook pro I don't take any offense from that comment. – RHELAdmin Jul 01 '10 at 14:28

8 Answers8

10

Others have already commented on the disadvantages. I'd like to say a few good things. Advantages to using laptops:

  • Power backup - assuming that the batteries still work, they would provide a sort of built-in UPS for the servers.
  • Lower power consumption - as you've already mentioned.
  • Built in KVM - no need for external monitors/keyboards, assuming the screens still work.

Your machines are more than enough to run web-servers and what nots. So, it is a non-issue. They may even be powerful enough to run as a VM host to become a mini-cluster of small VM machines.

Just a thought.

sybreon
  • 7,357
  • 1
  • 19
  • 19
  • Never thought about the little built-in UPS it has. Even though I'd probably only get an hour or 2 out of them with the screens off, it's still handy! – RHELAdmin Jul 01 '10 at 04:15
  • I used to have 3-4 web servers on laptop for this exact reason. (and cause i was poor and couldn't afford a rack or real servers) But they did the trick nicely. – grufftech Jul 01 '10 at 04:28
4

Disk will hurt you the most with a laptop, their disk IO is usually about half of an equivalent desktop. Also you don't say how many is in the Si unit of measure "bunch" - if bunch >= 20 then you could have lots of fun with things like MySQL NDB.

Geraint Jones
  • 2,483
  • 16
  • 19
  • Yea network database perked my interest as well. Theres only 8 at the moment (working) and a bunch of parts lying around that I could scrap into a few more. – RHELAdmin Jul 01 '10 at 03:33
3
* A web server
* A mail server
* A DNS server
* A file server

First concern is that these are old machines with limited I/O - so the best use you could put these to is in an application where nodes failing terminally and losing their data is very survivable - i.e. as webservers or DNS servers (LDAP servers too) where you are essentially just publishing static information.

If you are receiving data over the web then it does not undermine what is basically a good idea - just keep the data on a reliable shared substrate accessible from all the nodes in the cluster (e.g. a dedicated database or file server).

Edit:

If one of these devices has a probabilty of failure of 1%, and a new machine has a probablity of 0.1%, then two laptops in a cluster have a joint probability of failure of 0.01% - i.e. ten times as reliable as the new machine.

C.

symcbean
  • 19,931
  • 1
  • 29
  • 49
2

I think the main problem using laptop as server is about availability of the machines.
In a normal server the most used part is for sure the hard disk. In that case on normal servers you have RAID to save your data from a disk failure.
With a laptop, can you permit yourself to loose data in that case?
I think you could you those laptops for a not critical server as DNS, where data are not changed so often and in case of failure you could count on a secondary server.
Or a web server with a distributed configuration (drdb + heartbeat or openais).
Or for some development machines or to test some new configutaions before apply them to a production machine.

PiL
  • 1,591
  • 8
  • 6
2

Here's the core of what you get from a business-grade server, that you don't get with consumer grade machines:

  • Powerful RAID controller with high disk IOPS (we're talking 3mb/s vs 300mb/s)
  • Enterprise disks designed for 100% uptime and a high MTBF
  • Redundant power supply paths
  • Redundant/teamed NIC configurations with fail-over and load balancing
  • Remote hardware monitoring (SIM/OM)
  • Out-of-band management and remote console ability
  • Hot/cold aisle compliance and standardized form factor
  • Vendor Support

Assuming your requirements are low enough that none of the above is an issue, the final point to consider (IMO quite carefully) is the reflection it makes on yourself (and if you have one, your department) if you go ahead with this. While IT isn't always client-facing in the usual business sense, it's effectively a service to the rest of the business. How's the business take on running on unsupported configurations for these systems? How's it going to reflect on you professionally?

IMO the only time a hokey system like this may be appropriate is in an extremely small business where there's simply no cash and you're trying to bootstrap the whole thing. I'm talking mom-and-pop's flower shop and their kid is keeping their workstations, website and internet connection going. Nowadays, for anything larger than that, you'd be better cobbling together a few cloud-hosted services (dropbox/skydrive, gmail/yahoo) simply because they'll offer small-scale solutions with a solid infrastructure you don't even have to think about.

Of course all this is nonsense if you're just playing around with some servers at home. In which case some old laptops sound ideal.

Chris Thorpe
  • 9,903
  • 22
  • 32
  • -1: Its nothing to do with whether a single server can meet the demand - but whether the cluster can meet the demand. – symcbean Jul 01 '10 at 12:35
  • 1
    Google is built on commodity hardware. The whole can be more reliable than the parts. – Tobu Jul 03 '10 at 12:53
  • Guys, for starters you're talking about clustered software, and in the case of Google a highly customised stack built for a specific purpose. Second, meeting capacity is only one element of running an infrastructure, it is not the whole. – Chris Thorpe Jul 04 '10 at 00:54
1

Main factors that come to mind:

  • Heat dissipation
  • Power regulators
  • Hard drive capacity

Heat dissipation will be helped by the fact that they're in a cooled closet and well ventilated.

Power regulators on laptops often die (usually due to heat) and are soldered onto the motherboard. Not a fun thing to fix.

Unless there are extra drive bays (e.g. instead of optical drive) you're limited to 1 HDD, and you won't have hardware RAID. You could try a SAN/NAS but that is limited by your network card.

Laptops are a classic example of "planned obsolescence"; they break, wear out, or become outdated, and can't be easily fixed, upgraded, or repurposed.


A purpose for them... folding@home farm?

Andrew
  • 7,772
  • 3
  • 34
  • 43
  • I'm not worried about 1 drive. The file server won't be holding anything massive, mainly some music and lots of source code I've written over the years (possibly have this one doubling as a Git server) so 250G is plenty. Worse comes to worse I could make a SAN setup as you mentioned. – RHELAdmin Jul 01 '10 at 03:39
1

if they are not hit hard and you can run all the programs in memory, then they are fine, as long as they are well ventilated and the room is cool.

johnshen64
  • 5,747
  • 23
  • 17
0

http://www.rocksclusters.org/wordpress/ if you just want to play with the laptops.

MattrixHax
  • 141
  • 1
  • 3