Fog Creek Software
g
Discussion Board




Recommended servers for putting in colo

As I've mentioned the Joel on Software server lives in a colocation facility downtown run by Peer 1.

The server is a Dell PowerEdge. It comes with remote control hardware called DRAC III which can be used to remotely control the computer, reboot it, power cycle, etc.

The trouble is that the DRAC III is flaky as hell. The remote card seems to die all the time. The only way to regain control of the computer is to go to the physical computer and unplug it for a while. There is no physical way to reboot the DRAC without power cycling the whole computer.

Also, Dell's software skills leave much to be desired. The software and admin utilities associated with the DRAC III are terribly designed, badly written, and buggy as hell. Dell tech support has been of no help whatsoever.

So, here's a question for all you sysadmins out there: do other hardware vendors have better remote management hardware & software? Is this a common problem with Dell's DRAC systems or do I just have a lemon? Are other vendors better?

Joel Spolsky
Fog Creek Software
Wednesday, April 14, 2004

Joel,

I am not sure about the Dell part of this setup, however I have used APC power strips with remote management. You can log into the power strip and turn the power off to a certain machine and then turn it back on. This is as fancy as I have gotten with such setups, but it worked every time something went wrong at 4am.

Jeff

Jeff
Wednesday, April 14, 2004

Yeah, we have the power strips. The trouble is that the Dell gets itself into a state where when you power cycle there will be a message on the screen saying something moronic like

THE RAC IS HAVING PROBLEMS TODAY. PRESS CTRL+F11 TO BOOT ANYWAY.

and it will sit there for half an hour until I can get there. This is but one example of the incredibly bad software skills down at Dell.

Joel Spolsky
Fog Creek Software
Wednesday, April 14, 2004

For your purpose I guess anything from Sun is out of the question. So that leaves HP, IBM, and Compaq servers. Anyone of these major brands take their racks seriously? Small brands are welcome but I don't think it will be taken too seriously if the major brands are reasonably priced.

Li-fan Chen
Wednesday, April 14, 2004

Given the increase in the last five years of PC-class hardware used for servers, I'm surprised none of the vendors are taking remote management seriously.  On the Unix side of the world, Sun has Remote System Control, which basically lets you control the machine via a serial port, which can be connected to a terminal server, for example.  I'm sure the other vendors have similar facilities.

Hopefully, the PC server vendors will catch on...

joev
Wednesday, April 14, 2004

It's not that they don't take it seriously, and it's not that they don't have the features (what dell has is a lot more sophisticated than a serial port) -- it's that hardware companies are never any good at software, and Dell is especially bad, spending less money on R&D percentagewise than any other major hardware vendor...

Joel Spolsky
Fog Creek Software
Wednesday, April 14, 2004

You could try "PC Weasel 2000":

http://www.realweasel.com/

They've been around forever and have a good rep. What most impresses me about them is that they provide links to all their competitors to make it easy to compare:

http://www.realweasel.com/oph.html

Anonymous Coward
Wednesday, April 14, 2004

Joel, what features of the DRAC do you use other than remote reboot? Since the remote reboot of the DRAC doesn't seem to be sufficient, have you considered just ripping it out and going with the externally managed power strips? Obviously it's nice to have, but if the DRAC prevents the thing from starting up half the time, I'd think you'd be better off without it altogether.

(I took a quick look on Dell's web site, but I can't easily figure out what the DRAC does besides reboot)

Michael Kohne
Wednesday, April 14, 2004

Disclaimer:  I don't know anything about this product/company ;-) I just spent some time on google is all and found: http://www.peppercon.com/eric.html

GiorgioG
Wednesday, April 14, 2004

Have a look at www.raritan.com , their Dominion series of IP based remote control hardware.

Jason Watts
Wednesday, April 14, 2004

One recommendation is to stick to really boring, non-exotic, non-cutting edge hardware. Less moving parts means less problems.

Often people select servers with SCSI, RAID, Remote Controls, dual processors, hot swapping disks and fancy software like customized BIOSes thinking it would be more reliable. From my experience it generally isn't. While RAID 1 or 5 can improve system reliability if done well, it just as often makes things worse. Firstly, RAID adapters introduce another point of failure. Not a good thing, especially in PC type servers that often come with lower quality and less tested RAID hardware and remote controls.

A much more important factor is that exotic hardware is less likely to be tested sufficiently by BIOS, OS, kernel, driver software developers. The majority of hardware problems is caused by either cheap hardware (eg no brand memory, cheap power supply) or exotic hardware (RAID, dual CPU, etc). Aim at the middle.

I manage my own dedicated servers for 2 years now and I had to hard reboot exactly zero (=0) times. Luck is probably a major factor, but those Pentium III's and Seagate ATI harddrives sure don't give up easily. Neither do they give the software developers a hard time.

Also, when less experienced I would rather lease than colocate. What hardware do you trust more: your custom build server or the one that a good data center has running in 2000 fold?

Another consideration with colocated hardware is handling hardware failures. What happens if a harddisk or memory goes bad? When leasing a good data center swaps the faulty part in minutes. Or if everything broke they can put your harddrives in a similar other server. For your colocated box you have to go there and do it yourself. Even if you did provide spare parts/systems, the data center technician has to work with an unfamiliar system.

Another good thing about leasing is that you can change your server easily if you don't like it.

Jan Derk
Wednesday, April 14, 2004

Why would you do colo in the first place? I worked at a place for a few years where we just leased the equipment. The hosting facility knew the equipment inside and out and always had spare parts and all the few times we had a call because a server was down they were already fixing it. I really don't understand why people have to use their own computers. One poor sod was sold a DEC Alpha for US$27K (I saw the invoice) because he wanted a fast website. Holy Cow. And they stuck this in a facility where the power would go out when it rained heavy and they'd forget to turn that box back on and what a joke it was. rackspace or viaverio or someone like that.

Me
Wednesday, April 14, 2004

verio. . . lol.  Never worked for them I guess?

Elephant
Wednesday, April 14, 2004

"Dell is especially bad, spending less money on R&D percentagewise than any other major hardware vendor..."

That's because Dell is, in all reality, nothing more than a reseller.  They buy standard off the shelf parts and assemble them into a computer.  No need for that expensive R&D stuff that might cut into Michael Dell's fortune.

"Anyone but Dell" would be a good recommendation.

My Cousin Vinniwashtharam
Wednesday, April 14, 2004

Disclaimer:

If you're just running one server, this recommendation is not really relevant.

If you want amazing remote capabilities,  check out the IBM blade servers.  You get some seriously dense hardware, and the remote access is rock solid.  You can reboot the PCs, access the bios etc. all over TCP/IP.

DigitalEffects
Wednesday, April 14, 2004

Actually Sun is an option, if you don't mind installing windows yourself separately

http://www.sun.com/servers/entry/v20z/ has Lights Out Management

oh my aching wrists
Thursday, April 15, 2004

---""Anyone but Dell" would be a good recommendation. "----

My cousin Vinniewashtaram has evidently never bought a Compaq!

Stephen Jones
Thursday, April 15, 2004

*  Recent Topics

*  Fog Creek Home