Slashdot Mirror
NetWare 3.12 Server Taken Down After 16 Years of Continuous Duty
An anonymous reader writes "Ars Technica's Peter Bright reports on a Netware 3.12 server that has been decommissioned after over 16 years of continuous operation. The plug was pulled when noise from the server's hard drives become intolerable. From the article: 'It's September 23, 1996. It's a Monday. The Macarena is pumping out of the office radio, mid-way through its 14 week run at the top of the Billboard Hot 100, doing little to improve the usual Monday gloom...Sixteen and a half years later, INTEL's hard disks—a pair of full height 5.25 inch 800 MB Quantum SCSI devices—are making some disconcerting noises from their bearings, and you're tired of the complaints. It's time to turn off the old warhorse.'"
Netware 3 ruled.
Netmare 2 on the other hand earned the name.
By version 5 it was back to Netmare (for different reasons).
I once walked into a dusty environment, remote location and could hear the drive bearings from 100 feet away through a fire door. Backed up successfully but never spun up again.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
http://www.theregister.co.uk/2001/04/12/missing_novell_server_discovered_after/
Netware 3.12 was quite secure and rock solid. It did one thing (file and print serving) very, very well. It's a testament to good software design. The fact that you make light of it probably indicates that you were not in the IT field back then and have no sense of perspective. I wasn't a huge Netware fan, being more of an OS/2 and Unix guy back in the day, but I had a great deal of respect for the product.
Last year, I worked next to a system with DOS and IBM BASIC that has been up continuously on a production line since 1989, mind you, it was in a protective box with special filters and 90% of it's "functionality" is no longer used.
16 years and they did not run of space on it?
also good hardware not to fail in some way other that time. Did they hot swap UPS batteries over the years as well?
"My linux systems require constant patching for them not to be p0wned by script kiddies. Therefore it follows that every other system is the same.".
Love that logic.
Novell asked people to send screen shots of their uptimes. http://www.novell.com/coolsolutions/feature/103.html The winner then had an uptime of about 6 years.
You know those little squiggly red lines under words you type? I think they're trying to tell you something.
In Soviet Russia, dot slashes YOU!
So? Is there some rule requiring every tech website to report unique content?
I don't follow Arstechnica, so I'm glad that having been on Arstechnica doesn't disqualify something from being on slashdot.
From the linked thread: ... The only thing it's been connected to since 2004 has been my personal computer (laptop)."
"When I began work here in 2004, this system was completely orphaned
Way to spend (by my reckoning) 10,000 kWh of electricity.
Yup.
(1600 years old and thirty push-ups, but yup).
"I bless every day that I continue to live, for every day is pure profit."
I have very fond memories of 3.12, running a few site servers in the US, Mexico and Honduras. Although installing or patching was a pain with boxes of floppies to feed in to the server. It did what I asked it to do.
When I first looked at IPv6 addresses I had an IPX flashback. When we transitioned to IP from IPX (and to NT 4) I thought "these numbers seem finite compared to what is possible in IPX."
Whatever happened to the "Old Novell Guys" website from the late '90's? I am one. /Sorry if I am rambling.
Just a dude. Stuck in IT.
Essentially - other than tunneling IPX over TCP/IP, which the site may or may not have been using - this version of Netware had no TCP/IP support. No web server, no nothing. Odds are this this wasn't much of a risk. My guess (the article didn't say) is that they were using it for something really specific.
Put that old war horse down easy, it did it's duty and then some, it deserves some respect.
I loved Netware and worked on 2.x, 3.x and 4.x, it's a real shame what's become of Novell.
Is the order of magnitude these heads have traveled (in a circle).
8000rpm x 60 x 24 x 365 x 16 x (5.25/2/2)*pi x 12 x 5280
The special enclosures and controllers in server class hardware supported hot swap [e.g. rebuilding a RAID after disk failure]. The segments of a RAID array comprising a Volume could be on more than one disk controller. Really fancy systems had cold spare drives that would be spun up only when needed to rebuild the array.
Only volume sys: needs to be mounted at all times, other volumes could be mounted and dismounted by console commands. [And again server class hardware supported physically swapping the dismounted volume]
Caution: Do not stare into laser with remaining eye.
This was a NetWare 3.12 box and...
Hey KID! Yeah you, get the fuck off my lawn!
My guess (the article didn't say) is that they were using it for something really specific.
It's pretty obvious the only thing they were using it for was to watch the runtime go up and up.
The drive bearings had been noticeably failing for quite some time. The operators might pay some lip service as to why that somehow didn't matter, but the bottom line is - if the risk of a drive failure during operation isn't a problem, the machine isn't serving any real purpose.
#DeleteChrome
It's slower but more than fast enough, supports printers too although you'll really miss those Novell print queues. And Lantastic has evolved too, you are no longer limited to Arcnet, it supports the *new* 10baseT half duplex cards! Patches are available for the DOS stack to accommodate just about any combination of hardware IRQ and base IO PORT. Just be sure to load the network TSRs BEFORE you run Borland Sidekick.
Whoa! I was having 1984 flashbacks for a moment.
<blink>down the rabbit hole</blink>
Most of the patches I applied to 3.12 didn't require a reboot, and those that did didn't require shutting down the power. Such requirements as reboots are uncivilized.
deleting the extra space after periods so i can stay relevant, yeah.
tcpip.nlm
I think I first ran Apache on NetWare 4.11. 5.0 had a full stack, even the Tomcat server. GroupWise offered a web server on NW 4.x forward.
No, 3,12 didn't do all it could with TCP/IP, which was a little of a bummer.
deleting the extra space after periods so i can stay relevant, yeah.
That's a Montgomery Burns quote Homer.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Once a drive starts failing like that, the worst thing you can do is reboot the box... The drive may continue running for years, but if you shut it off it may never be able to spin up again.
Best thing is to get any important data off the drive without shutting it down.
http://spamdecoy.net - free throwaway anonymous email - avoid spam!
Kind of: http://en.wikipedia.org/wiki/Internetwork_Packet_Exchange
Just a dude. Stuck in IT.
This was on Arstechnica like 3 days ago. This site is increasingly feeding on news carrion.
This site has been doing that for years.
There's still no other site with the quality of exta information you get from the comments.
0 downtime per year. That's a perfect score, 16 years in a row. This is assuming the system was available all that time.
But if the network burpped once for 300 milliseconds, then it would only be eight-9s.
If you don't like the per year limitation of the calculation, let's say that the system was up for 16 years, but should have been up for 17 years (arbitrary). Then that's only one-9.
“Common sense is not so common.” — Voltaire
There was a patch for SuSE recently to fix a 208-day bug, where certain CPU registers on a particular CPU would hold their value through a reboot. We patched servers in work, and they still fell over at various times past 208 days of uptime. It was then realised that there was a need to fully cold-boot the affected servers for the condition to clear correctly.
So, that was one example of a patch that required shutting down the power.
- This sig deliberately left blank. Nothing to see, move along.
Oddly, I replaced my main home server with a highly energy efficient model four years ago (mac mini). I was using a kill-a-watt meter to measure that I was spending > $100/year on the old server, and that was a significant factor on what to get as a replacement. All my other systems are energy efficient laptops at home. I use the kill-a-watt regularly to test devices suspected of burning excess power.
Are there things I don't do? Of course. But I hardly ignore energy efficiency. I also make sure I'm not getting a low energy number that I will never make up the cost of over the life of the equipment. So that hybrid car? No go. I don't drive enough miles to justify the surcharge.
"I may disagree with what you say, but I will defend unto the death your right to say it." -- Voltaire
I always was rather impressed with those Quantum drives. I had a Quantum 1.2GB hdd in my computer when we suffered a house fire, and that drive was the only piece of electronics to survive in usable condition. Indeed, it lasted a good 4 or 5 years beyond that.
Dyolf Knip
I bought the 'spare parts' for the hard drive I most recently repaired on Ebay.
I had a failed 200g Maxtor drive. It had a lot of important stuff on it that I wanted back. It failed in such a fashion that it just quit spinning entirely so I gambled that it was an electrical problem on the logic board. I went on eBay and searched until I found exactly the same Maxtor drive, even down to the firmware version. It's nice that they have the zoom-able pictures on eBay and that many sellers post high resolution pictures.
It got me the data back perfectly. The repaired drive even works, though I'm leery about using it any longer.
Yes. It was called SPX (Sequenced Packet eXchange). You had an IPX address which is basically a MAC address and you preface that with a colon (:) and a six hex char SPX address.
The SPX address is roughly analogous to an IP subnet. The IPX address would be the individual IP host address. Note that the protocol doesn't stop you from repeating the full IPX range on each SPX network, you would have to override the automatic MAC address to IPX address assignment which would be unwieldy.
I suspect that a full modern internet running on IPX/SPX style addressing would look a little different from what we used back in the day. It might look rather more like IPv6 perhaps ...
Cheers
Jon
All of these replies about Novell Netware, and yet I haven't see one single mention of where Novell is today, how NDS came to be known as eDirectory, how Netware was ripped out and slapped on top of Linux under the name SuSe Enterprise Linux, which is totally free to download almost every product they ship and use on your own home network in an uncrippled fashion (so long as you don't want to security updates via a 30 day trial).
Anyways, cheers Novell, you will be missed o/ ;|
On Fark you get pictures.
Skip ------ See the latest from http://www.anArchyFortWorth.com
Only on slashdot you can laugh about silly comments, learn from really insightful ones AND learn about how HOSTS file can save your life and that APK posting as AC was impersonated by another Anonymous Coward!
Tomorrow is another day...
I heard a guy claim that if Novell registered network numbers as ICANN does today and insisted that every site had a unique network number then IPX might have ended up being the dominate supporting protocol of the World Wide Web instead of IP. But since 99% of the sites used network #1, you couldn't route IPX among companies.
This is a boring sig
IPX addresses had two parts - a 4-byte network number and a 6-byte host number that was almost always the MAC address. The network number was locally assigned, and in practice was almost always 00:00:00:00 (the default local network, because almost nobody actually bothered with routing), or FF:FF:FF:FF (broadcast), though some people got fancy and actually split up their networks into routed segments 1,2,3 etc. instead of bridging.
So you could theoretically run an Internet-like network on it if there were some central authority assigning network numbers instead of everybody rolling their own, and it would scale better than IPv4 because there were 32 bits of network number!
AT&T ran an IPX public internet in the mid/late 90s, in coordination with Novell. We assigned public network numbers, and sold connections. By now I've forgotten exactly what years it was, and I wasn't organizationally close enough to it to know if they actually got many customers, and of course there weren't really a lot of applications for it, but it probably ran for about two years.
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
Just a correction for JSG's post - the IPX address had two parts, a 32-bit network address and a 48-bit host address. SPX was separate - it's the Netware Layer 4 protocol that's roughly equivalent to TCP. IPX network addresses were locally administered, not globally, and most people just used the default network address of 0 (i.e. 00:00:00:00) and if they had multiple LANs they bridged them rather than routing, though some people got fancy and assigned network numbers 1,2,3, etc. the way they currently assign RFC1918 addresses themselves. The host address was almost always a MAC address (or broadcast.)
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
I guess I should be clear: IPX was layer 3.... SPX was layer 4. The use of the mac address in the IPX address reminded me of the use of the mac address in IPv6. The network address in the first part of the IPX address was the network. But you could assign any network address you wanted. So it had that shortfall.
Just a dude. Stuck in IT.
This actually makes me curious - in developing IPv6, how much of IPX ideas taken? One apparent drawback of IPX is that it was solely computer centric in that one would use the MAC addresses to formulate the IPX address. What if the device that needed an address (as is increasingly the case today) is not a laptop, but a phone, or a Bluetooth module, or something w/o an Ethernet card or Wi-Fi? IPv6 is good in that way.
One thing about autoconfiguration, though, at least in IPv6, is that it uses too many bits, and despite that, uniqueness is not automatically guaranteed w/o ND. I've argued a number of times that a smaller number, such as 32-bits, would have been just as useful, since networks ain't likely to have more than a billion nodes, and so within 32-bits, use any algorithm to create an auto-generated host address, while the network address is externally assigned. In fact, that would also have allowed for universal uniformity in network size, since all ISPs/organizations would have 64 bits, and the customers could decide whether they wanted more subnets or less and split the lower 64-bit space accordingly. Or assign 32-bits to the subnet, and the lowest 32-bits to the address. As for EUI 64, I agree that it's ugly, and what's worse - it enables any external IP spoofing tools to determine what the MAC address is, aside from wasting space.
Back to IPX. Since Novell was selling Netware, it could possibly have sold IPX network addresses as well, assigning them to every organization that bought them. Initially, Novell was restricted to DOS and Windows, but once they created UnixWare, they had the opportunity to popularize this w/ Unix as well. We would not be struggling w/ IPv4 address exhaustion today had this been done. Although once non-Ethernet devices came out connecting to the internet, Novell may well have had to change over to something like IPv6.