We've got two customers that colo laptops, for whatever reason. One's been there for about a year, the other for about three years (upgraded once during that time), and neither has had any hardware problem. Small sample size. YMMV. -Bill

I know it's not a laptop but, We recently started looking at some of the new Sun Netra T1's they are fast, 1u high, come in 110AC or 12 (maybe 24 I forget offhand) volts DC. They are real workhorses and perform great for the size and all. If you configure it with low memory (sun gets a lot of cash for their memory) they are pretty cost effective. I know that valinux has some small 2u boxes but don't know if they have any dc versions. Derrick I don't work for sun. In fact I would like to see a generic version of these I could run BSD on.

-----Original Message----- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu]On Behalf Of Jerry Scharf Sent: Thursday, June 15, 2000 1:08 AM To: Bennett Todd Cc: nanog@merit.edu Subject: Re: Laptops as servers?

One thing I know to watch for is periodic battery replacement. It turns out that running a laptop with a battery in it with AC on all the time will slowly erode the life of the battery. I have no way to know how long you can do this before the batteries won't do what you were expected. Maybe a battery replacement every couple years is all it takes.

The other is that the I/O on laptops sucks compared to PCI. For things that aren't major traffic handlers (like your DNS and DHCP examples) this is not a hit. Do any of the O/Ss out there get full rate out of a 100bT PC card?

jerry

(EE hat on..) There's an irony here. The desktop PC power supply takes the 120vac line, turns it at once into DC, then switches that to make high-frequency AC, runs the AC into a transformer and rectifies the outputs of it. (That way a tiny cheap high-frequency xformer can be used instead of a big heavy 60 hz one.) But for an almost inconsequential amount of extra effort, a different initial DC voltage could be used, say 48vdc, so that (CHEAP) battery backup could be added to any PC. (EE hat off) If they did, of course, APC, BEST etc would have a hard time. -- A host is a host from coast to coast.................wb8foz@nrk.com & no one will talk to a host that's close........[v].(301) 56-LINUX Unless the host (that isn't close).........................pob 1433 is busy, hung or dead....................................20915-1433

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 2000-06-15-01:07:44 Jerry Scharf:

One thing I know to watch for is periodic battery replacement. It turns out that running a laptop with a battery in it with AC on all the time will slowly erode the life of the battery.

That was definitely a major and severe problem at least a few years back; when the state of the art laptop used an NiCd battery, and had fairly dumb charging circuitry that easily overcharged --- and so overheated, and hence damaged --- the batteries, leaving it plugged in all the time killed it relatively quickly. People confused this with the subtle "memory effect", which they'd never be able to notice in laptop use. But a modern laptop, with brilliant power management software managing the health of a LiIon battery doesn't seem to suffer from this problem nearly as badly. I've been using my Sony Vaio 505TX for a year as my desktop computer; it's run pretty much a superset of 9-5 M-F plugged in to wall power, and I still get a reasonable battery run. And I'd still expect to use an external UPS with the laptop servers; I'd just expect the laptop's battery to provide a little extra coverage, and ability to ride over the switching. Plus it wouldn't load the external UPS --- or the air conditioning --- as hard. I honestly think that laptop designers and manufacturers have figured out people use their products as very compact desktop computers. In fact, if I'm recalling correctly, Toshiba pulled out an early commanding lead in the laptop industry in significant part because they had a _huge_ local market of office users whose office desks didn't have room for a US-styled "desktop" system.

The other is that the I/O on laptops sucks compared to PCI. For things that aren't major traffic handlers (like your DNS and DHCP examples) this is not a hit. Do any of the O/Ss out there get full rate out of a 100bT PC card?

As you say, for some jobs it's not an issue. For sure it's easy to get 10Mbps; I've done that for some time now with Linux through a 3C589. I haven't tried my 3C575 recently; a couple of Linuxes ago it solidly hung the system. But CardBus _claims_ to support data rates up to 132MB/sec. Even if it's hard to tweak a CardBus 100BaseT to run at full speed (I do not know), I honestly don't see that as being a major limitation for the kinds of applications to which a laptop would be otherwise suited. Maye it just reflects my youth and inexperience, but I tend to start thinking about higher-speed interfaces (faster than 10BaseT) about the same time I'm worrying about tuning a striped or raided disk farm with multiple controllers. A laptop with a single low-power (and hence low-rpm) IDE drive isn't competing at the high-performance end of things. There's actually a funny story there. I started using a laptop as a desktop computer before CardBus came around, and so got accustomed to the PCMCIA slot imposing a 4Mbps throttle on my ethernet. I couldn't figure out why people bothered with 100BaseT cards at all. I'd even confirmed the 4Mbps bottleneck with pathchar. Then I started with a new laptop setup, and was seeing real observed bandwidths up close to 10Mbps out of the likes of ftp, and was puzzled for a while. - -Bennett -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.0 (GNU/Linux) Comment: For info see http://www.gnupg.org iD8DBQE5SN5tL6KAps40sTYRAhjmAJ9EARos8O3ZGLLCMgcypgrJ9tqooACgjpjS L/qsSAY05nY41iTNHGOfQPY= =IV/B -----END PGP SIGNATURE-----

On Thu, 15 Jun 2000, David Charlap wrote:

IOW, it should be possible for a CardBus (but not PC Card) device in a CardBus-compatible PCMCIA socket to get equivalent performance to a PCI card. I have no clue, however, whether vendors are actually shipping devices that actually make use of all this.

ttcp between a 3c575 cardbus fast ether in my laptop and intel 82559 in a desktop (both linux) was 84Mb/s which is about what I'd expect.

-- David

-- -------------------------------------------------------------------------- Joel Jaeggli joelja@darkwing.uoregon.edu Academic User Services consult@gladstone.uoregon.edu PGP Key Fingerprint: 1DE9 8FCA 51FB 4195 B42A 9C32 A30D 121E -------------------------------------------------------------------------- It is clear that the arm of criticism cannot replace the criticism of arms. Karl Marx -- Introduction to the critique of Hegel's Philosophy of the right, 1843.

I had an "retired" Thinkpad running DNS. They definitely have less power comsumption that most other things, but I think the problem is mission critial redundancy. So they'd only be good for things like DNS and SMTP, that have fall backs built into the protocol, or for web servers clusters. It would be interesting to see a laptop with hotswappable motherboards, etc. -- James Smith, CCNA Chief Network/System Administrator DXSTORM.COM http://www.dxstorm.com/ DXSTORM Inc. 2395 Speakman Drive, Suite 2200 Mississauga, ON, L5K 1B3 CANADA Tel: 905-822-1957 (email preferred) Fax: 905-822-0680 1-877-DXSTORM (1-877-397-8676) "Feeling 'connected' with carbon-based dolts holds all the joy of being handcuffed to a dead zebra - it sounds special, but it can get old fast." - The Dilbert Principle - Engineers, Scientists, Programmers and Other Odd People On Wed, 14 Jun 2000, Bennett Todd wrote:

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

This whole rolling-blackout thing inspires me to ask: anybody tried using laptops as servers for core functions? I'm thinking infrastructure stuff like DNS and DHCP and Radius and so on, maybe backup SMTP, whatever else can be shoehorned into it. Who knows, maybe even Zebra one of these days. Logging servers. Firewalls.

Sure, if a server job absolutely requires a big farm of fast disks spread over a bunch of controllers, there's no point in looking at a laptop. But many server jobs don't. Enough to maybe make some blackouts hurt less? I'd _Sure_ rather provision UPS support for a rack full o' laptops than an equivalent amount of compute horses provisioned with traditional rackmount server gear.

They've got their own internal UPS, making 'em insensitive to switchover blips, and their running load is way less than normal server. 'Specially when their little lids are closed and their backlights and LCD drivers are all asleep:-).

Seems like it might even be worthwhile investigating the possibility of running 'em directly off e.g. 12VDC; I wouldn't be surprised if the standard power supplies for use in cars didn't end up being more efficient (spending less watts in the conversion) than the wall warts.

Hearing about the Google farm made me think, you know, it might be slicker to tuck 4 or maybe even 8 laptops into a U than just a pair of shallow rackmount servers. But then I'm weird:-).

- -Bennett -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.0 (GNU/Linux) Comment: For info see http://www.gnupg.org

iD8DBQE5SEWUL6KAps40sTYRAnjVAJ4qFNB6CKSujOD0id5NgoxEzFpGtgCgg4XS T6KK6V/pq6I1+WP08hWAeII= =tykJ -----END PGP SIGNATURE-----