LITTLEBLACKDOG.COM

Posted: Sun Jul 11, 2004 6:28 pm Post subject: Which CPU for Which Server?

imagen that we are planning to make our server and we are now talking about one part of the server which is the CPU

assumpitons:-

1- the mother board we have can only have one CPU
2- we have enough money to buy any CPU
3- the tempreture is fine for all CPUs under any load

and we want to by the:-

1- fastest CPU
2- and most relaible CPU with not down times

for sure, you will say cost is the most important field, because you can buy many cheap processors instead of buying one expensive and have a faster server. this is why i said that the mother board can only have one CPU. I dont want to talk about costs now, i hope to talk about the best with ignoring the "cost" factor.

as i read in Intel web page i found that the best proccessor in Intel family is Intel® Itanium® 2 Processor with 9MB L3 cache
[from: http://www.intel.com/products/roadmap/server.htm]

that was from Intel company. there are other companies than Intel that produces CPUs, but i dont know thier names to search about them.

I hope if you would tell me the best CPU from your experiance (not nessarly from Intel, it could be from any thing else). I know it could your opinion or feels about that CPU and not facts, but it is still important for me to know your opinions. And please tell me reasons why you like this more than that.

and finally, does the type of the service running on the server effect which CPU should I use?
does it matter if my server is HTTP or file server or Backup server to select the type of the processor?
is there any CPU that is the best for HTTP servers, and another one that is the best for file servers and so furth?

for example, the Itanium is Intel's best CPU product for servers. can i say that Itanium is the best for HTTP servers while xeon is the best for file servers?

Posted: Sun Jul 11, 2004 7:31 pm Post subject: Re: Which CPU for Which Server?

eech55 wrote:
as i read in Intel web page i found that the best proccessor in Intel family is Intel® Itanium® 2 Processor with 9MB L3 cache

If you say so. Shocked

Quote:
example, the Itanium is Intel's best CPU product for servers. can i say that Itanium is the best for HTTP servers while xeon is the best for file servers?

Itanium 2 is a 64 bit chip. You can't really make any general comparison. It is all based on what your company needs.

Mutt Joined: 09 Apr 2003 Posts: 101 Location: Middle Earth

Itanium 2's a crap, forget em.

You cannot look at just a single chip solution for a genuine server.

Dual chips allow for redundancy and higher thru put.

2 64 bit options i think you should consider are Opteron or G5 (IBM/Apple/Motorola 970 FX)

Both offer superior performance and relibility over the Itanium 2 and have excellent open source o/s options and custom Java solutions.

Posted: Mon Jul 12, 2004 6:56 am Post subject:

dropforge wrote:
Itanium 2's a crap, forget em.

The only "crap" about Itanium is that no one wants to rewrite software for it, and the x86 emulation (up until recently) was utter crap.

dropforge wrote:
You cannot look at just a single chip solution for a genuine server.

Dual chips allow for redundancy and higher thru put.

I fail to see how having 2 CPUs in one box will allow for redundancy when the whole box fails and goes under. Higher thru put, yes, but having even 128-node cluster will not save you from oh say, faulty power supply or corrupt memory, HD failure.

Intel chips generally score very well in integer performance, for a database server (MS SQL, MySQL, etc) that matters. For reliability, it's more in the chipset than the CPU itself. There's a myth that Intel chips are more stable than AMD chips, well... VIA makes the majority of the Noth/South bridges for Athlon mobos go figure. Chipsets made by AMD or Intel directly were almost always more stable. Xeon motherboards have been enjoying chipsets from ServerWorks and I do believe they're working on or have already made a server chipset for Opterons.

Guide Dog Joined: 08 Nov 2003 Posts: 8419 Location: MN

Lycander wrote:
dropforge wrote:
Itanium 2's a crap, forget em.

The only "crap" about Itanium is that no one wants to rewrite software for it, and the x86 emulation (up until recently) was utter crap.

dropforge wrote:
You cannot look at just a single chip solution for a genuine server.

Dual chips allow for redundancy and higher thru put.

I fail to see how having 2 CPUs in one box will allow for redundancy when the whole box fails and goes under. Higher thru put, yes, but having even 128-node cluster will not save you from oh say, faulty power supply or corrupt memory, HD failure.

that is why you have redundant power supplies, and mirrored raided drives, and maybe a raid 5 array for any important data you have stored... as for corrupt memory don't opterons have a memory stack for each cpu? now alas this won't save you from a motherboard failure..... but with a decent chip set (prefferable something that can be passivly cooled, to limit the possiblity of a fan breaking.) thos should be rare...

I can't say that daul cpu's has ever saved my ass, but I know redundant power supplies, hard drives, and fans have all saved me some down time, during peak hours... (still have to take it down to put a new drive power supplie etc in...) never had a cpu blow out just from use yet... only time I've had them burn up was if they weren't installed properly... (or more.. if the heat sink wasn't installed properly.)

Posted: Mon Jul 12, 2004 12:12 pm Post subject:

anglachel wrote:
as for corrupt memory don't opterons have a memory stack for each cpu?

AMD's SMP architecture has always had a point-to-point memory controller, so each CPU has its own path to RAM. A memory "stack" has more to do with software, each application has a stack, so if you have multiple running processes spread over multiple CPUs then each CPU would be accessing the stacks for the processes it owns. In other words it's nothing special. I was talking about memory corruption at the hardware level. The kind where you just have to yank the memory module out and replace it.

Posted: Mon Jul 12, 2004 12:49 pm Post subject:

anglachel wrote:
never had a cpu blow out just from use yet

I have... a bloody AMD. ran good for a year, then just pissed out one day. I've had IBM/Micron memory fail on me too... freaking gig stick at that too. Sad

You're questions eeech, while really quite specific, are hard to answer. Unless you subscribe to the "you get what what you pay for" theory. If that is the case, get a blank check, head to you computer superstore, and ask for "the most expensive ____________ you have."

Answering these questions is like answering "Mac or PC, which is better," "Linux or Windows, which is better." Everyone will have different answers for different reasons.

Personally, Intel CPU coupled with an Intel Mobo/Chipset will give me piece of mind at night. I've tried different combinations of memory, motherboards, and drives, and I'd sell my soul for an Intel CPU/Mobo, Micron/Crucial memory and maxtor or hitachi drives. Of all the pc's i've built or used... this combo gives me the best satisfaction for the money. Sure i can dump an ungodly amound on IBM drives.. but for the money, they are no better to me than Maxtor.

What CPU? What mobo? How much memory and drives? etc... depends on end use, and even there you have people who subscribe to different schools of thought.

Sad truth is you can run an HTTP server on an 1GhZ celeron with 512 mb RAM with no problems. If you're running only a few or just one domain, and it runs as you desire (response time and all)... does that qualify as "the best?" You can also dump a jillion dollars and get the latest and greatest , but with a poor network setup, or slow dataline... who cares if you're running an Itanium with 20 gigs of ram, and a U320 RAID-50 if your stupid dial-up account keeps your website visitors from viewing your site. I've gone to the extreme true... but you get my point?

Basically... i think you will find it hard to get one accurate and true answer... even when asking for the "best."

Mutt Joined: 09 Apr 2003 Posts: 101 Location: Middle Earth

Quote:
I fail to see how having 2 CPUs in one box will allow for redundancy when the whole box fails and goes under. Higher thru put, yes, but having even 128-node cluster will not save you from oh say, faulty power supply or corrupt memory, HD failure.

I was under the impression we were talking about servers.

Server CPU performance is measured by thru put and I believe that was the question.

The other parts are largly irrelevent as all have redundancy in a server, ie 2 of each, and can also be hot swappable. No need to power done to replace the faulty component.

The Opteron also has Lights Out Management, ie remote control a powered down server for restart, diagnostics and configuration.

Also, this forum has a few members who prefer alternative Open source o/s'.

The Xserve G5 would seem TOO perfect if Open Source compatibility was on the Agenda. They are looking at a few of them for where I work, unfortunately I won't get to plat with them.

Mutt Joined: 09 Apr 2003 Posts: 101 Location: Middle Earth

I forgot to put down why I said Itanium 2 was crap.

As was mentioned the architechure is different.

It may run Win2K3 but not .net, requires a lot more expense to recompile code, also more expensive to start with.

If IBM, HP and Sun are turning to the Opteron alternative you would need to consider it as well.

Moderator Joined: 26 May 2001 Posts: 8155 Location: Borneo

anglachel wrote:
Lycander wrote:
dropforge wrote:
Itanium 2's a crap, forget em.

The only "crap" about Itanium is that no one wants to rewrite software for it, and the x86 emulation (up until recently) was utter crap.

dropforge wrote:
You cannot look at just a single chip solution for a genuine server.

Dual chips allow for redundancy and higher thru put.

I fail to see how having 2 CPUs in one box will allow for redundancy when the whole box fails and goes under. Higher thru put, yes, but having even 128-node cluster will not save you from oh say, faulty power supply or corrupt memory, HD failure.

that is why you have redundant power supplies, and mirrored raided drives, and maybe a raid 5 array for any important data you have stored... as for corrupt memory don't opterons have a memory stack for each cpu? now alas this won't save you from a motherboard failure..... but with a decent chip set (prefferable something that can be passivly cooled, to limit the possiblity of a fan breaking.) thos should be rare...

If you really spend the money you can even have redundant mobo's. You can even have a computersystem that is running on two places at ones and will run on if you loose one of the places

Posted: Mon Jul 12, 2004 6:28 pm Post subject:

dropforge wrote:
I was under the impression we were talking about servers.

We are

dropforge wrote:
Server CPU performance is measured by thru put and I believe that was the question.

He also wanted to know about stability and reliability.

dropforge wrote:
The other parts are largly irrelevent as all have redundancy in a server, ie 2 of each, and can also be hot swappable. No need to power done to replace the faulty component.

I don't doubt that you know this, and I certainly am not questioning your knowledge of servers, I meant no disrespect. I was simply picking on your statement about having 2 CPUs in one machine - an SMP machine - for redundency. Redundency - in my mind - is a fail-safe couter measure, a backup. Having multiple CPUs working to provide more throughput is simply a performance measurement, BUT it does not imply reliability nor high availability.

dropforge wrote:
Also, this forum has a few members who prefer alternative Open source o/s'.

I be one of them too.

Mutt Joined: 09 Apr 2003 Posts: 101 Location: Middle Earth

Quote:
I don't doubt that you know this, and I certainly am not questioning your knowledge of servers, I meant no disrespect. I was simply picking on your statement about having 2 CPUs in one machine - an SMP machine - for redundency. Redundency - in my mind - is a fail-safe couter measure, a backup. Having multiple CPUs working to provide more throughput is simply a performance measurement, BUT it does not imply reliability nor high availability.

No Disrespect taken (or given to you), I lept to the assumption reliability meant only 1 thing, a 24-7 server, which means the built in redundancy of doubling up.
If one cpu drops the other cpu takes up the work until a replacement is sorted and installed with no loss of service. A single cpu would mean the server is down.

I did not think past my opinion of the unsuitability of a single cpu server.

linky [/url]http://www.apple.com/xserve/[url][/url]

Guide Dog Joined: 08 Nov 2003 Posts: 8419 Location: MN

Lycander wrote:
anglachel wrote:
as for corrupt memory don't opterons have a memory stack for each cpu?

AMD's SMP architecture has always had a point-to-point memory controller, so each CPU has its own path to RAM. A memory "stack" has more to do with software, each application has a stack, so if you have multiple running processes spread over multiple CPUs then each CPU would be accessing the stacks for the processes it owns. In other words it's nothing special. I was talking about memory corruption at the hardware level. The kind where you just have to yank the memory module out and replace it.

fine then stack was the wrong word... (though not entirely, but mahaps there was a better choice) still if you have memory for cpu 1 and memory for cpu 2 that is one stick you can have go bad ... and possibly not take you down in the mean time... and if the mobo supports it replacing it well still running (though it sounds unlikely to me I wouldn't be supprised at all if it was either already out there or already in the works at amd or intel labs...)

Posted: Tue Jul 13, 2004 9:06 am Post subject:

dropforge wrote:
If one cpu drops the other cpu takes up the work until a replacement is sorted and installed with no loss of service. A single cpu would mean the server is down.

Whatever the cause may be that would take out one of the CPUs, I'm guessing it would be for electrical or heating problems. If it were electrical than that might well screw up the whole circuitry and render the whole system moot anyways. CPUs aren't plug and play, they're detected upon bootup so if one were to be removed or break during normal operation, the OS would have a fit.

anglachel: same goes for memory, the modules are detected at bootup and the OS reads that figure from BIOS (or maybe certain Assembly instructions). I don't think the OS would like it very much if a RAM module crapped out in the middle of operation. Yes, plug & play RAM would be awesome, they'd sell like hot cakes. Makes it easier to diagnos (ugh I can't for the life of me spell that word right now) problems and much faster if all it is, is bad RAM.

Mutt Joined: 09 Apr 2003 Posts: 101 Location: Middle Earth

Quote:
Whatever the cause may be that would take out one of the CPUs, I'm guessing it would be for electrical or heating problems. If it were electrical than that might well screw up the whole circuitry and render the whole system moot anyways. CPUs aren't plug and play, they're detected upon bootup so if one were to be removed or break during normal operation, the OS would have a fit.

anglachel: same goes for memory, the modules are detected at bootup and the OS reads that figure from BIOS (or maybe certain Assembly instructions). I don't think the OS would like it very much if a RAM module crapped out in the middle of operation. Yes, plug & play RAM would be awesome, they'd sell like hot cakes. Makes it easier to diagnos (ugh I can't for the life of me spell that word right now) problems and much faster if all it is, is bad RAM.

No dude, Server O/s' (like win2k3) know when hardware is faulting and can do the load balancing, page the admin and if necessary power down the server. The O/s can also detect what is hot swappable and what is not.

My first experience with this was swapping 2 HDD's in a Dell server. First you let the server know what you are up to, then when the spare disk was built, yank out the dead unit, put in the new, make sure it builds then off I go. Most times the Server has already built the spare and a little orange led tells you which has died, yank the drive, shove in the new one, server automatically detects it and sets it up as the new spare.

One of the seniors informed it was even more fun hot swapping a cpu in front of the customer. People find it hard to believe that stuff is possible.

I am prety sure all the Opteron servers these days have hot swappable cpu and RAM modules and that Microsoft and others write their server software to recognise this.