Ok all you Sun/Unix experts...

I need a very rough ballpark figure for hardware to support the following (I know these arent really enough to spec a box on - but have a go) btw. I'm neither a Sun/Unix or a Database expert so feel free to call be a moron

1 Box to Support Oracle DB. likely 200Gb for the database. Regular large updates - but from one client - which is anlysing log files. Ad hoc queries but we're not talking lots of clients all querying away all day. One or two clients with occasional query.

Its mainly an info store rather than an on-line style system.

If the database had to grow to a)500Gb or b) 1000gb what are the implications ?

Any ballpark style figures appreciated.

Or looking at it another way - if you had 10K to spend before the year end - could you make a start on the hardware for the above (maybe leaving out the disks ?)

Deano

 

If you want a rackserver (including Sun Sparc) have a look at
www.rackserver.com which is run by a friend of mine who's been doing it for years.


Jeff

 

I can sell you a nice competetive solution via Fujitsu Siemens.

All too often people forget that Sparc isn't just Sun..

Where about are you based? I could organise a nice demo of the Fujitsu Siemens kit in Bracknell if you wanted to see whats on offer.

Fujitsu Siemens have a large advantage over Sun in that their machines are more efficient thus you pay less on your per processor licencing from Oracle.

Pop me an email with as much detail as you want/can and I will get you a quote together. We can even do equipement rental including regular upgrades if you wish.

 

YHM to your hotmail account

 

ok..

The question is not only the size of the database, but the amount of clients that have to connect to the database. A small server will support a 200 gig database quite easily, but if you want to scale to 1TB you will have to do a lot more that just "plug in another 50 disks"

With databases that size you have to plan the layout of the data across the disks to cope with the large / sequential read + writes that the database will have to support.

The sun servers that are the bees knees at the moment are the V880's. They can store 12 internal disks, totalling 320 GIG storage + 2 mirror'd internal drives.Sun have a special on them at the moment.. On the other hand, the favorite oracle databases are E4500's that you can pick up quite cheap. Couple that with a load of D1000's / A3500 (ot T3's if you have the money) and you can scale from 200gig to 5-8 TB on one machine!

The good thing is that they support 8 cpu's as well..

Fujitsu stuff is OK if you're not really serious about SUN and Solaris, and you have never had to work on both platforms.. (joke ).. But seriously, compare the datacentre space dedicated to fujitsu, then the space dedicated to SUN. Bank's don't make that mistake twice!

My 2p is that if I had to run a sun application on a SPARC / solaris server, it would have to be SUN. There are so much problems with cross ownership of problems if you have one supplier for the disks, one supplier for the OS and one supplier for the hardware - support can be a nightmare!

 

The question is not only the size of the database, but the amount of clients that have to connect to the database. A small server will support a 200 gig database quite easily, but if you want to scale to 1TB you will have to do a lot more that just "plug in another 50 disks"

Yep wholly agree with you...

With databases that size you have to plan the layout of the data across the disks to cope with the large / sequential read + writes that the database will have to support.

Yep...

The sun servers that are the bees knees at the moment are the V880's. They can store 12 internal disks, totalling 320 GIG storage + 2 mirror'd internal drives.Sun have a special on them at the moment.. On the other hand, the favorite oracle databases are E4500's that you can pick up quite cheap. Couple that with a load of D1000's / A3500 (ot T3's if you have the money) and you can scale from 200gig to 5-8 TB on one machine!

Yes they are nice machines we are also a Sun reseller and yes there are some good deals to be had on the E4500 however scaling is an issue and you can't go to 128 processors

The good thing is that they support 8 cpu's as well..

Fujitsu stuff is OK if you're not really serious about SUN and Solaris, and you have never had to work on both platforms.. (joke )..

I will take you joke as that

But seriously, compare the datacentre space dedicated to fujitsu, then the space dedicated to SUN. Bank's don't make that mistake twice!

erm two major UK insurance co's use FSC kit over Sun. Yes the banks don't use it YET FSC's domain model for SPARC processors far outruns some of the Sun's spec. The abillity to roll over parts of apps rather than whole apps as well, have you looked at FSC kit recently? there isn't the space disadvantage there used to be... - and I agree with there USED to be a disadvantage..
If you want to have a look at the kit there is a fellow scoobynet member who runs the FSC demo centre and would be quite happy to show you the ins and outs of it...

My 2p is that if I had to run a sun application on a SPARC / solaris server, it would have to be SUN. There are so much problems with cross ownership of problems if you have one supplier for the disks, one supplier for the OS and one supplier for the hardware - support can be a nightmare!

Err not wanting to cause a huge debate here but I along with FSC or Sun can supply you with an solution that is a one stop solution.

I suggest FSC are currently buying market sector and thus the price advantage is there.

 

Thanks for the Mail - And all the comments here.

Been overtaken by events now - Accountants keep changing their mind about how much we can spend

My recommendation amongst overthings was that we shouldn't do anything in a rush -'cos we're 80% certain to get it wrong.

All our existing hardware is Sun Solaris - and I'm not in a posistion to affect a policy change in hardware vendors though I can always point out that there are alternatives.

We will be doing the database later in this year - so I'm more that a little interested in the pros and cons of the last couple of posts. As I said the server hardware and support isnt my area but I am part of a small team so like to know at least a little (enough to be dangerous ) about the other aspects.

Deano

 

We've just implemented a v880 for a client , replacing a 420R previously used (this is a webserver btw). Both were 4 processor, 4GB ram machines and with some software tweaks and 2 hardware crypto/SSL cards it took 150% growth over last year without breaking a sweat.

Nice kit. Get's my vote.

stevem

 

deano... yhm.

David

 

Are the FSC boxes based on and old USII design with a _few_ extra bits. Guess if they are it would be like buying a Pentium 1Ghz when you can get a 2Ghz P4 for almost the same money.

Edited to add...

Yeap I know clock rate is not everything but my understanding is the USIII cpus are almost twice as quick at the same clock rate as USII chips. The FSC cpus did have a few percent over the USII cpus if I remember correctly.

[Edited by orbv - 3/13/2002 12:42:06 AM]

 

The FSC servers use SPARC64 GP processors and a Fujitsu-designed motherboard/backplane. They are 100% compatible with Sun Solaris. They are a totally different design to Sun machines and are in no way a clone. They are made on the same production line that has been making mainframes for over 25 years. Dont forget that as well as Sun, Fujitsu are part of the SPARC group that actually develpoed the processor.

As WillieF says, a major insurance company use them over Sun (now have over 45 servers), and one of the MAJOR banks in the City now has 5 and is buying more as we speak. There are other machines that I personally know about in other banks in the City aswell (I was involved in putting them there ).

One thing worth noting about the V880 is that they run hot. I think something to do with passing the cooling air over the PSUs before the system boards. I have known Datacentres that have complained about this.

The FSC servers are the size they are for a reason - they are built to mainframe class.

Edited to say that the FSC machines typically run slower clock speeds than Sun's (helps to keep them cooler) but they consistently run quicker overall - 4 clock cycles for every 3 on a Sun.

Dave

P.S. Yes I do run the FSC UK Demo Centre as part of my job !!


[Edited by druddle - 3/13/2002 11:30:17 AM]

 

The FSC high-end stuff may be mainframe class, but sun has been catching up with the mainframe's as well over the past few years.
I don't think that sun have released any specs on the E15K yet, but with 106 900MHZ cpu's and half a terabite of ram it's not going to me much of a slouch but the real bonus is linear scalibility, due to the crossbar interconnct..

it's just nice to be able to discuss pro's and cons without being flaimed for having an opinion

 

I'm interested in the space deabte...

One of our problems is a reluctance to use much rack space. We are a private network with POPs all over the UK. We are limited therefore in our choice of hosting environments - which means we pay too much for rackspace - Especially in the data centre these would be going in.

As always its a toss up between cramming as much in ad risk the air flow not being suffcient or buy more rack space.

Deano

 

The FSC high end stuff is mainframe class, you can see by the build and the size of the ruddy things !! But the midrange machines are also built to the same class on the same production lines. On the midrange stuff the cooling is done front-to-back and so is perfect for rack mounting. Access is also through mainly the back, but you can get in to some components thru the front (unlike some Sun rack mounted servers where access is thru the sides). Because the systems have more PSUs than they need, the PSUs run cooler as there are more of them to supply the current. Also as i mentioned, the clock speeds are slower so the CPUs run cooler.

I agree that a E15k with 106 900MHz CPUs and 1/2 terabyte RAM would be very quick, but in order to go up to 106 CPUs you lose I/O slots. Thats not the case with the Primepower 2000 - it can go up to 128 CPUs /512Gb RAM (4 per system board) and you can still have 6 PCI I/O cards (fibre, dual/quad ethernet, scsi, etc) per system board so you have 192 PCI slots for the machine. Also with a fully configured Primepower 2000 you get around 57Gb/sec data transfer rate across the backplane and thats not slow !! And the Primepowers have been using the crossbar interconnect for the last 2 1/2 years

For more Primepower performance info look at the following 2 links.

http://www.fujitsu-siemens.com/power/Benchmark.html

http://primepower.fujitsu.com/en/

I have not had a great deal of experience with Sun, but I can talk all day about the Primepowers. There is a large install base in Japan, Germany, France, Benelux, the US and the user base is growing steadily in the UK. I know of some big Sun houses that are testing Primepowers at the moment to replace such things as E10k, 6500s, etc.

Also, Fujitsu Siemens in Germany have a Solaris development team that actually write code with and for Sun that is included in Solaris. You get exactly the same "Blue Box" to install a Primepower as you do a Sun.

Dave

(Edited cos i cant spell)


[Edited by druddle - 3/13/2002 11:22:32 AM]

 

Mmmmmm.....

FJ Servers Vs Sun Servers
Skoda Cars Vs VW/Audi

IMO, thats where the problem lies

I will continue to buy Sun Servers because Im a snob

 

ok then,

not that I am a geek or anything

BUT


>>The FSC high end stuff is mainframe class, you can see by the >>build and the size of the ruddy things
The're massive! The E15K is about the size of the smallest one of the enterprise systems (PrimePower800)

>>On the midrange stuff the cooling is done front-to-back and so is perfect for rack mounting
All of the current sun rackmount servers (apart from the desktop workstations that can be rackmounted) cool front to back. Period.

>>Access is also through mainly the back, but you can get in to >>some components thru the front (unlike some Sun rack mounted >>servers where access is thru the sides)
The only servers which this is true is the E450 / V880, but these support hot-swop of the PCI cards. The rest have all major components on boards (or blades) that can be hot-swapped out if needed, front or back.

>>Because the systems have more PSUs than they need, the PSUs run >>cooler as there are more of them to supply the current. Also as >>i mentioned, the clock speeds are slower so the CPUs run cooler
Every system should have more PSU's than it needs. If they don't, I'd fire the sysadmin .. CPU's running cooler is not really an issue in any enterprise datacenter...


>>Thats not the case with the Primepower 2000 - it can go up to >>128 CPUs /512Gb RAM (4 per system board) and you can still have >>6 PCI I/O cards (fibre, dual/quad ethernet, scsi, etc) per >>system board so you have 192 PCI slots for the machine
It takes up the floorspace of 4 E15K's! No wonder it has 192 PCI slots!


>>And the Primepowers have been using the crossbar interconnect >>for the last 2 1/2 years
First pioneered in the E10K, back in.. oh.. 1997??

'fraid that I can't comment on the performance of the E15K, as no benchmarks have surfaced to support any arguments that I may start - so I think that I had better shut up there!

Edited to add that I love both skoda and audi/vw, and as long as it runs unix it had gotta be a good thing!


[Edited by krankyd - 3/13/2002 12:18:26 PM]

 

Well yes i will agree that the Primepower Enterprise servers are large and weigh shed loads. But during some earth tremours in Japan a while ago, in a certain area it knocked out all the servers in the building, with some of them even having components shaken out. All except the Primepowers.

As for the PSU statement, you are right kranky, every server should have redundant PSUs. The Primepower can be installed with a secondary AC inlet, which means that on an 800/1000 you can have up to 12 PSUs for the machine, and the machine would require 6 x 32amp single phase power supplies !! All six AC supplies are used, but only a small amount of their capacity and so run cooler, which can benefit the server room environment. Thus if you lose the whole system power supply (4 x 32A) then it will still run on the power of the auxillary unit. Also, CPUs running cooler can have the same environmental effect for PSUs, but also _can_ increase the reliability and MTBF of the units. I think the Primepowers (from stats of all systems in the world and the faults they have had) have been calculated as having a 77000 hour MTBF time (8.7 years).

Fujitsu have been making the crossbar backplane in the mainframe and Enterprise servers for over 25 years. And yes the E10k used a crossbar too. At the end of the day the Sun server must be good, as they have got a massive install base in this country alone, but one of our biggest customers came up with the statement that they think that Primepowers are "better Sun than Sun".

Anyway, as you say as long we are dealing with servers running UNIX then the world will be a better place

 

cool - at least we can agree to disagree

speaking of earthquakes, I remember when I worked at exodus, and the sales guy was showing us aroung the cages for the first time.
He was proudly saying that the building / power / cages were build to withstand earthquakes up to 5 on the richter scale.. This was fantastic, as we were in west london, earthquake capital of the world!

 

This just makes the intel architechture look ****.. suppose could go compaq himalaya (speling??) range or IBM Numa-Q, but I'm sure these are RISC based....

incidently we use a couple of netra-t1's an enterprise 420 (old..) and IBM RS6000's