SQL server hardware

[BigV]

I have been tasked to produce a plan for a system to run our SQL databases. This will be a replacement system for the one that flinched this morning.

Some requirements:

First tier hardware, including asap service contracts

Redundant systems so that there is no single point of failure, including the server itself

Windows Server OS

(not directly a server for sql requirement, but some accommodation for backup must be included in the plan)

Less crucial details:

Other Windows networking roles will be assigned to this newest, bestest system

Rackmount preferred, plenty of space >20U

Technologies I'm willing to consider at this large junction:

SAN / NAS

Server virtualization

----

What is working for youse guys? What is *not working* that you would warn me to keep away from? I have a nice situation here to be able to design what I want. I want to learn from your experiences, please.

[smoothmoniker]

I have an old dell desktop running Ubuntu in my closet that I use to download porn and mp3s - I can just host it on that. Cool?

[BigV]

And the score going into the bottom of the first inning:

Definite hardware specifications -- 0
Experiences to stay away from -- 1

I love this game!

[Perry Winkle]

I'm not familiar with "first tier." I googled it and it seems to mean "top name-brand." Is this right? Does an ASAP service contract mean that you have someone on call at the vendor?

Are co-los or 3rd-party providers options?

[BigV]

First tier means top name brand. HP, Dell, etc. Someone who makes premium gear *and* has the service organization to respond very quickly to support requests, mostly of a hardware nature.

I don't want a colo for this. This is strictly in house equipment.

[Undertoad]

At my workplace we do the big DB stuff with Oracle on HP running Linux. We have two massive HP disk arrays, and HP and Dell servers accessing them. My experience so far is that HP is utterly expensive but service is good; while Dell will dispatch contractors who know the Dell gear but are not quite as motivated. The HP arrays ("EVAs") are beautiful, and so easy to configure that it's hard to do it wrong. A single array contains multiple switches and controllers so the whole thing is just a massive cluster. No s. p. of f. out of the box.

Also, Oracle clustering is a bitch to understand from scratch.

[SteveDallas]

Originally Posted by BigV

I have been tasked

You do realize that in modern corporate-speak, "tasked" is synonymous with "bent over and fucked hard, with no lubricant," right? Just sayin'.

Originally Posted by BigV

Redundant systems so that there is no single point of failure, including the server itself

Originally Posted by BigV

I don't want a colo for this. This is strictly in house equipment.

This whole project is out of my league, but these last two statements may be contradictory. If you really are looking at 100% up-time, then you really need to consider a failover mirror at a remote location.

I would suggest googling around for some of the recent stuff about disk reliability that Google has come up with in running their own data centers. (It's too late on Friday for me to find primary cites myself.) I believe the bottom line is they're starting to come to the conclusion that it's more reliable to buy dirt cheap systems and mirror the hell out of them (I'm talking 3 or 4 mirrors here) rather than put a shitload of money into some fancy-pants SAN that becomes a single point of failure. But again, this is out of my league.

You said nothing about networking or bandwidth requirements... what's the prognosis for a server that's up and running, and the network switch it's connected to barfs?

Originally Posted by Undertoad

At my workplace we do the big DB stuff with Oracle on HP running Linux.

I'm not even going to touch the one about "no single point of failure" and "Windows OS"!

[BigV]

UT: Thank you, sir. I, too, have found HP to be the gold standard, and that is reflected in the price. But compared to the cost of the absence of this kind of concierge support, it is a fair deal for the purpose these machines serve. The way things are shaping up around here is HP in the server room, and Dell on the floor and in the field. It's working so far.

My main problem is that I buy servers so infrequently that I have to relearn how and what each time I get some. I rely heavily on the experience of others' experiences; others whom I trust, like you, like SD, etc.

Sorry, what is "s. p. of f."?

SD: Tasked, asked, aksed, axed, assed, arsed... it's all the same. Though I am keen to avoid being axed. "Tasked" was my choice of words, though my boss did say get me a quote. It's a task. Piper payers call the tunes...

Actually, after some additional conversations with the boss, a colo might be in order. Our main sql app vendor offers a hosted solution. It is definitely in the realm of possible answers to this problem.

As for single point of failure... yeah. Well. No matter what disaster I prepare for, there is some disaster + 1 that is more than I bargained for. I don't have the chance to rebuild everything from scratch, but for those things I do get to replace/rebuild/design I try to put in as much headroom and redundancy as I can justify.

We recently deployed a *small* Citrix farm--two servers. But, that choice, made at the outset, makes scaling up in the future much easier. I'm thinking along the same lines with this sql project. Each piece I install I do so with a mind to the future and what traps I'm laying for myself.

I have not had a switch barf and kill my server's accessibility. It was the ups powering the switch. I know.. I know. Roseanne Roseannadanna was an IT prophet.

As to the Windows OS... It's my long suit. It may be an ugly suit, but it pays the bills. Steady work, though, you must admit.

PW: Good service for me means 4 hour response time. I don't need a vendor rep onsite or anything quite that exotic.

I will say that HP has embarrassed themselves today. The HP server failed, actually, three of the six hot swappable fans in the case failed, and I asked to have (under the terms of my 4 hour response contract) the replacement fans sent to me immediately. Four hours later they told me it would be another two hours.

Eventually, the courier showed up. With two fans.

I installed them, and one of the new fans was DOA. Actually, the hot swap bay is dead, because switching the fans in the bays shows that the fan's ok, but the bay's dead. Great. Another call to HP. Yes. I need a technician this time. The verdict is the mainboard. Great, roll'em. Now. It will be dinner from the vending machine again tonight.

Now, just now, the server's down again. Shit.

[Undertoad]

single point of failure

[BigV]

well, don't I feel like a penny waiting for change...

[Cicero]

Can I bring up a database for fun?
Off topic- this database is fun for home use. You can entirely customize it quite easily. Just for fun.......Well I had fun with it- didn't want to buy it at the time. Neat and easy to play with. That is all....

Sorry V- I suck.

http://www.download.com/Database-Oas...-10568283.html

[BigV]

After a few calls to HP's hotline in Costa Rica, I finally reached someone who said, yes, I see your case number. No, I don't know why they told you that the technician would be calling you.... this is coded for next business day service, and that means Tuesday.

:sound of head exploding:

Well, I have comforted the poverty of their ignorance. The ticket has been recoded to immediate or whatever their little checkbox on the screen says... The technician called me from the road. He's about 30 miles north of here. Says he'll be here in about an hour. That's optimistic given rush hour traffic, and rain and my bad luck today.

Plan C is to buy a new server at the store tomorrow and build one from the ground up. Sigh.

[Undertoad]

They don't have enterprise-class servers at a store!

[smoothmoniker]

my offer still stands. I'll even add an external USB drive for backup storage, if you need extra SPOFs, or whatever.

[mbpark]

#1. You want to get an HP Proliant. You really do. There is a reason they are the top Windows-based server in the enterprise.

#2. Don't run other additional roles for this box at all! That's a big security faux pas. You can buy cheap DL360's to run IIS or Active Directory on.

#3. If you have the budget, get the DL580 G5. I just built a quad 2.4 Ghz quad-core that will be getting 16GB RAM (it has 4 now, other 12 is getting shipped since my customer buys the RAM separately due to price). We're getting another one with 24GB for reports.

#4. Run 64-bit Windows Server 2003 R2 on it. This way you won't have to tweak memory settings to get more than 3GB allocated to SQL Server.

#5. Run SQL Server 2005 EE 64-bit. The performance between 32-bit and 64-bit is night and day.

#6. Look at getting a low-level Fibre Channel SAN. The HP servers are standardizing on 2.5" Serial Attached SCSI drives. It's better to just get a FC SAN and run a few LUNs for your data, logs, and backups to the server. This way, if you want to cluster, it won't be hard to do. iSCSI is not as mature as FC (4Gbit/second over fiber), and HP offers great deals on Storageworks equipment that can handle it. This way you can just set up your LUNS as RAID 0+1 for data and log, and RAID 5 for your backups and transaction log backups. HP also builds great stuff here.

#7. On that note, you can fit all of your OS + Application needs on 4 * 72 or 4* 146 SAS drives in a RAID 5 + Advanced Data Guard (ADG) configuration. The on-board RAID controller on the Proliants ships with this.

If you follow some semblance of these notes, you will have yourself a very high-performance DB.

Mitch