Fog Creek Software
Discussion Board




Who dropped the ball: Peer1, Dell or FCS?

It would be interesting in light of recent discussions here to know who was responsible for the JoS outage yesterday. The site was down for most (all) of the Europe business day.

Just me (Sir to you)
Wednesday, February 19, 2003

Funny that, someone else said something similar yesterday and the thread disappeared overnight! spooky!

But yes, I noticed same

Gwyn
Wednesday, February 19, 2003

In one of the threads Joel mentioned that this site would soon be moved to peer1, maybe that's what happened.

Prakash S
Wednesday, February 19, 2003

Yeah, but a thread about the site being down has been deleted, by someone in at Fog Creek I am guessing. I wonder why they thought they had to delete the thead?

Ben
Wednesday, February 19, 2003

I too, on more than one occassion, have asked about the moderation (cough cough censorship) on these boards and not recieved a single response on why certain posts/threads were deleted.

Joel, can you shed any light??

tapiwa
Wednesday, February 19, 2003

Questions about moderation are not allowed. This thread will self-destruct in 1 hour. ;)

Jan Derk
Wednesday, February 19, 2003

I would imagine Joel's gonna be pretty embarrassed about yesterday's outage considering he was was spieling (or should that be spoeling?!) forth with how the new setup was going to be oh so resilient. Perhaps that $35 switch failed after all!

Gwyn
Wednesday, February 19, 2003

Let's be clear: this is not about embarrassing Joel or anything. It is just that this is a significant event in light of it being discussed so much in the last weeks, and maybe we can all learn from it.
Personally I can not see how it would be worthy of less bits than the mike story.

Just me (Sir to you)
Wednesday, February 19, 2003

I too noticed that the JoS outtage thread has been removed and it is disheartening to say the least.

Jon Newton
Wednesday, February 19, 2003

The operating system vendor could have dropped the ball too.

Ryan
Wednesday, February 19, 2003

Unfortunately, it's not at all that exciting.  I was moving some stuff over to the new server and I rebooted it, but there was something in the BIOS that was asking for someone to press F1 to continue on the console.

I had to go down to the colo facility to figure it out.

Michael H. Pryor
Wednesday, February 19, 2003

Must have been a hell of a traffic jam to get there!

Gwyn
Wednesday, February 19, 2003

Thanks for the reply Michael. Appreciated.
So, it seems the lesson for us all is:
When colocating equipment, make sure you can do EVERYTHING you expect to do remotely before you leave the colo facility.
OTOH: Since Peer1 does offer remote hands and feet 24/7 could you not ask to have them have a look at the boot for you?

Just me (Sir to you)
Wednesday, February 19, 2003

The Peer1 guys were *extremely* helpful.  They did remote boot it for me, but when it didn't come back up I went down because I thought something serious was wrong.

Michael H. Pryor
Wednesday, February 19, 2003

These conspiracy theories are so ridiculous. Why are people so quick to jump on the bashing bandwagon?

pb
Wednesday, February 19, 2003

We still do not know what the moderation policy is on the boards

tapiwa
Wednesday, February 19, 2003

at least you know why JoS was down:-)

Prakash S
Wednesday, February 19, 2003

Interestingly JoS was unavailable again for a while with the 'SQL Server' not found error. Ho hum.

Gwyn
Wednesday, February 19, 2003

Gwyn, get some work done, will ya:-)

Prakash S
Wednesday, February 19, 2003

<quote>
the BIOS that was asking for someone to press F1 to continue on the console.
</quote>

Aaaah. That must have been the "No keyboard present. Press F1 to continue." error.

Jan Derk
Wednesday, February 19, 2003

Get a sun, configure it to be headless then it doesn't care about a keyboard.  Intel junk doesn't like to go headless.

Cedric
Wednesday, February 19, 2003

Yeah, it's pretty hard to disable the keyboard error message on that Intel junk.

What sucks even more is how well cheap Intel hardware works for a small site like this.

Joel had better be on the phone with Sun right now!

d00d
Wednesday, February 19, 2003

<<
Yeah, it's pretty hard to disable the keyboard error message on that Intel junk.
>>

I think you can tell the BIOS not to halt on errors.  That *Should* resolve the issue.  I could certainly be wrong. 

GiorgioG
Wednesday, February 19, 2003

I belive you can purchase keyboard and mouse terminators to plug into the ps 2 ports.  But a Sun box is especially nice because it doesn't suffer from this stupidity from the get go.

Cedric, proud owner of a sunfire 280
Wednesday, February 19, 2003

Ahum. Hello everybody: I was joking. Or at least trying to...

Any current Intel/AMD computer will boot without a keyboard.

Jan Derk
Wednesday, February 19, 2003

Oooh, a Sunfire 280.  At least you bought an UltraSPARC generation without silent data corruption issues that Sun tried to cover up.

Rodger Donaldson
Wednesday, February 19, 2003

I know a 280 is pretty small by Sun standards, but it will walk on any 4 way xeon box for database work, you see MHZ is kind of a lie.

Cedric
Thursday, February 20, 2003

Hey Cedric,

have I found a deal for you. Listen, I happened to stumble upon these great timeshare condos in Arizona ....

Just me (Sir to you)
Thursday, February 20, 2003

Just me.  Stay in your intel world.  If you see how computing really works, it would scare you

Cedric
Thursday, February 20, 2003

As you seem so knowledgable about this issue I would love to hear your comments on

http://www.realworldtech.com/page.cfm?AID=RWT012603224711

Just me (Sir to you)
Thursday, February 20, 2003

Processor This has so many gigahertz,  Intels newest baby runs circles around the older UltraSparc, anyone surprised?

Processor speeds is pretty much secondary to the other stuff that affects performance when building a system, disk I/O speeds, bandwith of memory buses, cache implementations (cache miss penalties can be severe).

With Suns and other more expensive machines a more "well balanced" system in all parts is what motivates the considerably higher price, aside from a cool logo :-)

You can pretty much forget about this or that many Mhz.

Have fun,

Patrik
Thursday, February 20, 2003

If you are intrested reading a book on processor and system architecture, and what design considerations you should make i can reccomend reading:

Computer Organization & Design
The Hardware/Software Interface
written by David A Pattersson and John L. Hennessy.
ISBN: 1-55860-282-2

Patrik
Thursday, February 20, 2003

>Oooh,  At least you bought an UltraSPARC generation
>without silent data corruption issues that Sun tried to
>cover up.

Roger, are you saying my ULTRA5 (UltraSparc IIi) has silent data corruption problems?

Patrik
Thursday, February 20, 2003

Just me,

I am speaking about currently widely available kit.  Sun has the deck stacked agains them for a few years.  Maybe they can pull off a big win with Sparc V, maybe not.  IBM is strong in the game.  The Itanium might be a good processor.  Who is going to make chipsets?  Intel?  Laughable.  HP is going to make their own, so probably will Fujitsu.

As Patric stated.  The whole architecture is more important than, but often overshadowed by processor speed.  Definitely the case with current Intel 32 bit kit. 

Also, Windows is a team of chickens hooked up to the wagon, wheras UNIX is the draft horse.

Cedric
Thursday, February 20, 2003

Cedric,

there is quite a bit of Sun kit around here, also some Compaq (Alpha's) and even some old SGI's.
Guess what: 90% of the new stuff coming in is Intel based, and I am not hearing complaints.

Just me (Sir to you)
Friday, February 21, 2003

Just me,

>Guess what: 90% of the new stuff coming in is Intel
>based, and I am not hearing complaints

Guess what: That is totally appropriate. 90% of the stuff where I work is Intel machines as well. In my experience some people use Intel based servers for the 10% of jobs they are ill-suited for as well.

We have some major DB systems where I work, and for "heavy-duty" DB stuff we have Oracle and DB2. The UNIX based server I work on is a 8 way Sun 4500 machine
and what we use it for, Intel hardware would probably not pull off with the same performance.

Again, Intel PC are not well balanced systems, A PC is very responsive at first, but they almost die under heavy load, wheras well balanced systems can keep up with a constantly heavy workload without dying.

Patrik
Friday, February 21, 2003

Patrik,

To put it in car terms.  Intel have the highest rpm, but don't have low end torque. 

Intel does work for a lot of things, just not those that are under a heavy load 24/7

Cedric
Friday, February 21, 2003

Gentlemen,

Intel based systems come in all shapes and sizes. Some are pieces of total junk, with a Ferari engine bolted onto a Sinclair C5, others are top of the line systems balanced as a F1 car. You can expect more varience since this is an open platform.
Although I am not generally a great fan of benchmarks,
as an example I think http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103022001 can be still be educational for you.

Just me (Sir to you)
Friday, February 21, 2003

Sir,

System Information Total System Cost 5,619,528 US $ 

That is a moderately priced piece of iron.

I know you can buy and build well balanced systems with Intel hardware. Hey, you can even get a Beowulf system,
thing is you can not buy it pre-built from Dell or whereever.

Patrik
Friday, February 21, 2003

No, but you can from NEC or UNISYS or Compaq.

It is hard to predict since Sun has long since been shamed out of running any app level benchmarks, but my guess is that a equivalently configured (ram/disks) Dell 2650 based system (the same stuff running this board) would keep up quite wel with a 280R, at about half the price.

Just me (Sir to you)
Friday, February 21, 2003

Sir,

>No, but you can from NEC or UNISYS or Compaq.

I stand corrected. I thought these were custom builds from NEC, one-offs for really demanding clients. Not something sitting around on shelves at NEC or Compaq.

Patrik
Friday, February 21, 2003

Patrick,

Odly enoug a 4500 (albeith with 14 processors where you have "only" 8) is the last Sun to be found on TPC http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=101092502 at 1,734,522 US $ 

Oddly enough it is passed by a significant margin by a 4 processor Intel machine http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=102121601 at 441,022 US $ 

hmmm ... but maybe that is not fair. After all, the Sun was available in summer 01, let's see ...

Ah, what about this Dell http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=101082301 from around the same period? Just as fast on 8 processors (PIII Xeon at the time) and all that for 591,071 US $

Maybe you need some Sunshades ;-)

Just me (Sir to you)
Friday, February 21, 2003

Then again, I dont run something as obscure as
SymfoWARE Server Enterp. Ed. VLM 3.0
on mine, and my shades are one of them cool 80's mirror Ray Bans.

I'll keep them, and the Sun too ;-)

Patrik
Friday, February 21, 2003

Patrik, Cool, you got the prison guard/bossman with the shotgun shades.

Just Me,
Unisys makes some very nice kit and I suspect Fujitsu will as well. 

TPC benchmarks are fine.  Windows does well on the TPC-C because usually they scale horzontally, but with the Unisys they can go up.  If you want to see where Wintel fails look at the TPC-H benchmarks.  Wintel is noted for its complete absence beyond the 300Gb benchmarks. 

I believe the biggest reason for this is lack of 64 bit capability.  Windows and and Intel only now have this.  Large databases love ram and the 4 gig limit or the fake ram addressed with AWE are not the same as real ram sitting their for caching query results, stored procs etc.  It will be interesting to see if Wintel can start showing up in the TPC-H space

Cedric
Friday, February 21, 2003

Patrik: Potentially, yes.  Although I didn't realise that the US III has this, as well: http://www.theregister.co.uk/content/2/18102.html ; see also http://www.theregister.co.uk/content/archive/17417.html (if you don't trust the Register, you can get the same details from more traditional reportage).

The worst aspects of the US II bug was that Sun did their best to cover up some reasonably serious problems on their highest end systems.

Rodger Donaldson
Sunday, February 23, 2003

Patrick,

I agree SymfoWARE does not have the same familiar ring as DB2, Oracle or SQL server, but then again, there might be a reason why they are still top dog http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=101082801 .
My bet is Oracle would be slower on your kit (all benchmark caveats apply of course).
As for sticking with Sun, you might have very good reasons, but a blunt "Intel machines just can't cope" seems straying a bit too far from reality.

Cedric,

I especially avoided including any clustered machines in this comparison. While they might give you a significant boost, it does not apply to all scenarios and would at least require very substantial retooling unless designing new apps from scratch.
As for the large version (10.000 GB) of TCP-H, both machines there are clusters. Guess what the very top dog runs on: Intel.
As for the Microsoft references: Intel does not equate to running Windows or SQL Server. Still it will be very interesting to see the launches of Windows Server 2K3 and SQL Server 2K3.

Just me (Sir to you)
Monday, February 24, 2003

*  Recent Topics

*  Fog Creek Home