Wursti offline

[wurst]

Oki then let me tell u how the BAD on this Planet works:

First of it all we have Microsoft. They bought a System called Navision some years ago.
The System is what they call ERP, its a Software which is very flexible, its there for Enterprises to organize the whole Firma.
It was a great System, for M$ ist was worth 1,6 Billion Dollars i think.
Since Microsoft pwns this System, the Name changed several times.
Everything went inperformant, at the moment theres two Branches/Forks of that Software.
So far, so good. Who buys it, is an Idiot.

Theres some like a golden Rule for buying Microsoft Products:
DONT BUY THE THINGS THAT CHANGE NAME FREQUENTLY
They use to cover crap by this Mechanism.
Do u know Echange? Its good right? Well, do u know how it was called in the 90ies? Right. Exchange.
Office? Same Thing. Great piece of Software, the Name will never change.
And if: Be warned.

Lets say now: There where i work, some Idiot bought it.
Now I must live with it and btw: I must care for teh People that work with it.
Im a Microsoft Expert, 90% of my Work i do is Windows.
What dont mean that im a Computer Expert.
And sure it dont mean that 90% of the Machines that i run are Microsoft. Maybe 40-50. Hah.
I just know lots of their Products and how to get them running stable anyways.
Over the years i became a Master of dirty fixes, Drive Mirroring Technologies and else Workarounds.

Now we come to HP.
Microsoft itself only gives Support, if a Coder that they chose customized it.
And only certified Hardware Vendors are allowed.
If You break the rule, you will lack Suppport.
It will die one day and it will be Your fault. Good bye.
Sure it will hang on certified Hardware too and nobody will help u.
But it wont be your fault, u can stay with your Job.
Nice expectations.

Theres a choice now of three Hardware Distributors.
All of them play the same game: cartel.

Now u might say: Huh, cartel. Thats forbidden. Sounds a bit like The Godfahter I II and III.
Lets send a lawyer and all will be fine, penalties on this can easily exceeed a billion Euros.
Sure destroying capitalism is a serious crime in capitalism, heh.
Not as easy as that. Why?
Cause HP sits in Eastern Europe, some country where they have different Laws bout the destructive side of capitalism.
IBM is the second "certified", dont ask me for the third.
They all have the same in common: The Enterprise resides "somewhere" in eastern Europe.

How i got proove?
Hehe.
Due to the mentioned crappyness of the System I need expensive Hardware.
In such cases, u go to many HP authorized Dealers and request an offer.
I got suspicious immideatly, when i saw that those dealers offers vary for just 0,3%
I mean: If they would give me a Price from the List, it should be exactly the same.
But soooooo slightly different? Funny.
However. Thats no proove, right? Right.

It happened now, that nobody could deliver a specified Part, namently the P812 Controller.
So while phoning with one Dealer, i told him:

"Wow, no Problem. Just make me the Offer including the Controller.
If i cant have it in time (for my project), ill "fish" it somewhere from the Internet.
We can get bit more Ram then or a Client PC, your Salery will have minimum the specified Value or more."

"Oh", he said, "But that will cause another three days to wait"

Me: "W00t? Is ur Chef in a Holiday?"

He: "No, Sir. I must send the request to HP somewhere in eastern Europe. They make the Price what were supposed to offer"

OMG.
So i told him thx, ill take IBM.
But Magic: Same again there.
Aditionally, IBM is even more Expensive.
With HP u pay allready more then the double price, compared to the market.

Later, he told me that he could anyways do some.
"Ill tell them HP that u bought another Brand before, and that u are thinking about to switch to HP."
24 Hours later i had an Offer. Wow. 35% off, what a Discount!

And the others? They all rang me next day, telling me that a "great Discount" is now possible for specially me.
Cause im sooo-oh-hooo special.
U guess what im telling u now: Right.
0,3% Difference again between the dealers offers.

Anyways, i bought HP since im a little Admin ad i have no chance.
I got the fat Controller in time and ran a test with 6 SSD (OCZ Vertex2 Pro)
After succeeding with that test, i bought another 14 Drives.
50 GB each, in order to increase Performance.
My Target storage Amount are 500GB, im running Raid10 cause i knwo that no Raid5/6 Engine could do the Job.
It means that im losing half of the Capacity but however.

Alltogether it was thought to march straight forward for 500.000 IOPS, the Controller Specs promised me at least 4GB/sec.
Really nerdish Value, and, if needed for a 100-Employees Firma: A proove of being helpless.

Btw: This kind of Storage is not the future.
Its just a step inbetween. SSD have (as i mentioned) a specified amount of (re)write- Cycles.
The World is waiting for MRam. Production is starting these Days, in 5 Years that Storage will be obsolete.
Imagine a System that only have the classical reboot procedure for the case of Faillure or Hardware Changes. Great? Sure its another nerdish dream.
The SSD Disks that i used have special techniques implemented in the drive controller which increases the life time.
But one Day they will anyways tell me good bye.
But i dont have to fear that Day. First of it, i have Raid. Second reason is the appplication that im running there.
A Database Server is an ideal Environment for such a constellation since you can Backup every single Update by a "Transaction Log".
Webservers (which have very low writing and high reading expectation) might be another good Way to use them.

However, whoever drop in from teh google and read this:
Take it as a Warning.
Microsoft seem more easy at the first glance but its not.
You will end with ur feet in a bucket, someone will fill concrete inside and let the water raise to your Throat.

Then, my friend, is payment day.
Many other payment days will follow.
Scaling will be an old dream, you will forget bout it or cry your whole Life.
Whatever step your buisiness may go, they will get cash.
And You, Your employees and your business will go straight forward to hell.

[natirips]

I so understand how you feel.

[wurst]

Its so much money that get spent on freaked out greedy people.
normal people get born, educated, work and die for this pile of shit.
and i must buy it, im feeling like an asshole.

i didnt mean to do like this when i started networking.
I wanted to make the machines help and connect the humans. and not vice versa.

[SvaRoX]

35% discount woot oO
What alternative then ? These companies are built on the same model after all...

Any idea about the release date of the first MRam "hard drive" ? After UrT 4.2 ? =)

i didnt mean to do like this when i started networking.
I wanted to make the machines help and connect the humans. and not vice versa.

Nice

[wurst]

hehe, i think the conveyor belts spit 32 mbit chips since 2 month or so...
I think it will need another 2-3 years till first drives arrive on the market.
Whats not solved yet within MRam is a fitting interface, sure operating systems must change in order to use these drives (no booting anylonger).
Pci/pcie slots are the fastest "peripheral" connectors, theres no way to boot from such a slot directly as far as i know.
Not every comp is a database server (which need only a drive letter / mount) in order to profit.
Sure a mini conventional boot with early mount could be a workaround.

@this controller troubles:
I mean the HP lies here fatass, saying that 4GB sec would pass the inteface.
at least the 1GB cache should respond with interface speed.
however, if u buy adaptec/3ware/tekram etc: it wont pass the 2GBsec barrier.
They are just honest bout the speeds, thats the point where HP fails.
From this point of view, HP is simply not made for the use of professionals.
Toys with a shiny description, nothing on what u can count.

One way to archieve faster speeds is to use multiple controllers that can cluster (the HP ones btw cant).
Another idea is to run a mixed soft-hard raid10 (many little raid1 on many controllers, set as a big raid0 in operating system).
The win softraid is known as somewhat CPU intensive and a database server "likes to be alone at home"
However, when i make a Win Soft Raid0 from all those drives, im getting exactly the same speeds.
CPUs (we have 16+ HT) are idleidleidle, theres not 1 core eaten by the Win Softraid.
In *nix Environments it should be performant too. Linux softraid is known to be usable, eating allmost nothing in Raid1-Mode...

Ah, btw: The machine is at the moment faster then any other comparable Server.
Developers are very satisfied, they say its a Rocket.
We have only 3 users migrated by now, its the accountants.
They used a separate system before (which ate csv data), so they are easy (and the first) to be migrated.
The rest will follow during the next months, then we will see the truth.

The mixed hard1-soft0(virtual10)-Raid is an option that will work out for the worst case, im sure about that.
And yes: Fuck the bandwidth. Its about to handle a microsoft sql server on a (fragmenting) NTFS file system.
So its the IOPS that i need.
Theres nothing in the Specifications but however i turn it: The dead end with that P812 is 107.000 IOPS read and a bit less write.
A comparable Adaptec Controller would give 250.000, two of them would beat the shit out of that disks.
It would ctrl-alt-delete the hardware support, so its a no go.
The Servers have enough PCIe Slots, dont ask me for the Southbridge of this Intel 5000 Series Chipsets.
However, 100K IOPS is what u get from 500 SAS Drives. Pretty speedy.

Everything will be fine, i think the guys that decided (they are no IT specialists at all) learned their lesson.
I give this System 5-10 years, after there will be (hopefully) a concept that is basing on GPL.
Thanks to SQLang the Migration will me much more easy then it is today.
The funny thing is:
When u take a look at MS Axapta, u see something that is very similar to Internet Explorer 8.
Theres an Application Server between the DB and the Clients, its called AOS (Axapta Object Server)
Like the PHP is between you and the Mysql in our "little" DSWP case. Using Ajax, everything should be possible to be rewritten.
Thx to TsLib ODBC Drivers theres performant mostly Bugfree MSSQL Access from *nix Systems.
Let me call it "Light at the end of the tunnel"

However, Admin is finished for the first, others may work now.
m looking straight forward to read all 1200 Posts that were made during my little absence.
Im Sorry that i have no bigger Benchmarks for u, but it would be just benchmarking the Lazy HP Controller Bottleneck Speed.
If someone has different Controllers for testing: Its possible, SSD are free on Weekends.

And sure i learned some bout life, business and last but not least i compared a bunch of expensive SSDs which made my heart beating.
I think we will profit from them very soon on B3 Database Server.
They are not only fast, they save energy and have a MBTF that makes me sleep alot better.

[natirips]

Whats not solved yet within MRam is a fitting interface, sure operating systems must change in order to use these drives (no booting anylonger).

I think no booting wouldn't need much adaptation, it would be like stand by or pause in virtual machines instead of shutdown. However, a machine shoud still have some quasi-booting to re-establish network connections and the likes. I've experienced problems with network after resuming from stand by or resuming a saved virtual machine on many systems. I feel like more adaptation would be required on the BIOS level.

On another notice, I think Windows will gain a lot of speed with faster HDD, judging from the amount of HDD LED's typical flickering.

[wurst]

In my dreams, i would imagine it like that:
Some installations, error terror and setup would run the classic way...
Maybe they will add separate memory dump slots, where u can continue with your os where u once prepared it.
However, i dont know much about thse plans (if there are any...)

Ah, Mram is more slow then actual memory.
They describe it "as fast as sdram" (similar speed as RAM from 10 years ago)
Maybe the memory dump wit occour first, then the drives.

I think the more impotrant fact is, that industry dont spread ssd in a wide range.
some server distributors have them in program, but they lay "in the edge where nobody can see them"
They all hope for Mram, thats my opinion about this issue.

Sure web providers got them. But i think they dont buy IBM or HP. They dont need hardware support since they run clusters and ye, theres nobody to support debian. Hah.
Aditionally, the market is much harder then for an Enterprise IT.
They cant afford ineffective Hardware since they wont survive it.
Tyan or Supermicro are more their brands, they are cheaper and more bleeding edge. Plus theres driver Support that they need.
Hetzner for example use much MSI, i heard about Asus. Ah, I myself have an Asrock in 1 Server. Runns great, have a second one for savety...
And they use every drive that they can get since theres multiple copies.
ssd is a great option in that case. calculate:

Setup:
lets say we have average access 20KB on a webserver.
ah, and sure we have cheap gigabit network.

We need ...hmhm ... moment... ~6350 IOPS to fill the wire. Lets say 6000, its crappy ethernet.
So we get HDDs. Fast ones, WD Velociraptor 2,5" 15K SAS. 200 IOPS each.
Thats hmhm... 30 Drives. Each is 300 euros, so we spend 9.000 euros.
Aditionally we will need an expensive SAN (external refrigerator-like disk array) Cost for that box is some between 2K and 4K Euros.
Now the SSD Calculation. We get 1 SSD. It makes 20.000-50.000 IOPS. It costs 200-500 Euros. Finito.

When it comes to electricity, we compare 300 Watt for the HDD with 2 Watt for the SSD.

Sure the amount of storage will be "slightly different".
Even on those extra-expensive HDD, we have ~150GB for 300 Euros or 0,50 per GB.
SSD cost some between 3 and 8 Euros per GB, depending which chip type they contain.

So for raw backup store, the HDD is cheaper. For large Files, it dont make a big difference.
For DB and web applications, the SSD rocks.

The DSWP Page contains maybe 2MB "small" Files.
Then theres ~10 MB Images, its icons and minimaps.
Then we have maybe 500MB attachment Store and theres 1-2-3 GB Maps.
If u have the thing under control, u can make very nice Configurations from HDD, SSD and RAM.
Mass web providers dont do so atm, at least i didnt see anyone offer some like "10GB HDD / 2GB SSD / 256 Ram for web-self-mount"
If u own a PC: Great oppurtunities to gain some performance per Euro.
Lets call it tunining with brain, not with overclocking.

The little map pics here for example are RAM-cached, they reside in a (way too big) 32MB Ramdrive.
For RAM storage specially cached files are great, they will re-appear automatically in worst case.
Map Images will do the same, GDlib cares for them. The large original images are stored on HDD.

Even the Map download could easily fit inside some GB ram, ~30 Euros each.
With some brain it should be possible to just load one map in download server: The actual one. It would even help stop abuse.
Sure a Desktop HDD can serve those large 2-50 GB files very performant and cost effective. Even 20 GB aint a real bis fish.
So this stays as it is.
Gameclients get served atm by a 640GB Hitachi SATA desktop drive. 13 Cent per GB, 50 T€uro TCO, finito.