13 TB on Windows / SQL Server

Story: Nextaction dumps Microsoft for Oracle and open sourceTotal Replies: 9
Author Content
moopst

May 30, 2006
7:35 PM EDT
That's gotta be some kind of record. Not sure if "glutton for punishment" or "adrenalin junkie" is quite right.
grouch

May 31, 2006
6:36 AM EDT
It probably slipped up on them. You know you could start a recycling service by hauling empty aluminum beer cans in a toy wagon. You might not notice the shortcomings of using a toy until the business grows and the wheels start falling off from the load.
tuxchick2

May 31, 2006
7:14 AM EDT
It was 13TB on eight databases on five windoze boxen. Which is still pushing the poor things far beyond their limits.
dinotrac

May 31, 2006
10:21 AM EDT
tc:

That arrangement reminded my of my old days as a capacity planner for HUGE datacenters (as in EDS center devoted to General Motors) ...

and having to explain the real dynamics of the "all my eggs in one basket" bromide.

It is very good to spread out your risk, but there's a gotcha:

When you can't run on most of the eggs: ie, every egg matters, spreading them out actually increases your vulnerability.

Why? More things to break down. Not only that, using more things often means using less capabile things as the supply of money tends not to scale linearly with the things desired.

Now...using five boxen instead of two or three means you've got that many more parts, etc to fail. And, of course, they're Windows boxen, so they will.
tuxchick2

May 31, 2006
10:37 AM EDT
dino, you made my day. That's always been my favorite approach- minimize the number of widgets, because as you say, that creates more points of failure. And it makes administration a nightmare, especially on a horrible platform like windows, which can barely single-task, let alone handle a large distributed database. (MS-SQL server isn't that bad, I suspect that if it were ported to a real OS it would perform ably.) (If that's even possible, as it is probably infested with all manner of OS-specific hooks and horrid microsoft foundation classes.)

I have beat my head on the wall I don't know how many times over this issue. For some reason PHBs are fine with authorizing purchases of multiple cheap PC servers that you know are going to keel over, but not a single robust server made of actual server components. And even when you don't have dingbats in charge of purchasing, there's always the geek chorus gumming up the works by singing "I can do that on a 486 cluster!"
dcparris

May 31, 2006
12:17 PM EDT
WinXP Pro crashed when I was running MS Word 2000 - by itself, within the first week the OS was installed - and you mean to tell me someone would want me to run a mission-critical database app on something like that?
dinotrac

May 31, 2006
12:24 PM EDT
tc -

And, to be clear (to others: it's obvious that you understand the dynamic), we are talking about a specific (very common) situation:

When all of the pieces matter.

Google, on the other hand, has architected a fault-tolerant system via heavy replication -- lots of interchangeable, replaceable parts. That's a different paradigm altogether -- and they do budget in time and resources for constant replacement of failed servers.
tuxchick2

May 31, 2006
12:40 PM EDT
dino, to Microsoft "cluster" is the first two syllables of a three-syllable world.

dc, you are so right. I about died laughing when I got a PR packet on the MS Cluster Server 2003. One of the main principles of a cluster is protection from hardware failure- WTF are you going to do when the cluster software itself is the main point of failure?
moopst

May 31, 2006
8:35 PM EDT
I got to visit the fabrication building at Lockheed where they put together Atlas rockets. They use some engines made in Russia which have fewer parts than the American engines originally designed for the Atlas. The American engines had a separate pump for fuel and the hydraulic system that points the engine. Since the fuel is basically super purified diesel, the Russians used the fuel itself as the hydraulic fluid and so they only needed the one pump. Given that if a fuel pump or hydraulic pump fails the rocket is doomed anyway I immediately saw the benefit in terms of reliability.
dinotrac

May 31, 2006
9:05 PM EDT
moopst -

It's amazing how many people have trouble with that concept when it's applied to computers systems. Funny thing is, these days, computer systems can be more complex than rockets.

Posting in this forum is limited to members of the group: [ForumMods, SITEADMINS, MEMBERS.]

Becoming a member of LXer is easy and free. Join Us!