With the arrival of the AMD Opteron and Intel Itanium, commodity servers built on these processors have joined proprietary RISC systems from IBM, Sun Microsystems, Hewlett-Packard, and others in the 64-bit landscape. With prices starting at just over $2,000, Opteron and Itanium systems — running Linux or Windows — are already carving out a niche in high-performance computing clusters, where they are used to run compute-intensive scientific- and financial-modeling applications. Eventually they will replace their 32-bit forebears in corporate datacenters, and clusters of them may even challenge 64-bit Unix systems costing hundreds of thousands of dollars. More at Infoworld.
Here is a quote from the article:
On the downside, the power button is tiny and hard to press, and the keyboard and mouse ports were not labeled.
The guy is talking about a 1U Linux server that is usually used headless and powered 24/7.
<sarcasm>Very relevant, no?</sarcasm>
OK. I wont.
Infoworld apparently has a problem with other sites linking to their articles. When you click the link to the article, you get a page that says you can’t read the article because it was linked from a site that is “not authorized to redistribute our material” and they need to “protect their intellectual property”. WTF? When did linking to a web page become redistribution of the content?
Until somebody hits them with the clue stick, you’ll have to copy and paste the URL in your browser’s address bar.
It works fine here, with Mozilla 1.4.
I just *cannot* believe this guy reviewed the latest Zeus release against an Apache release which is a major version behind ! Zeus gets its speed mostly by being multi-threaded, whereas Apache 1.X is most definitely not.
No shock then, that this bloke bitches and whines about the performance of Apache 1.X vs. Zeus, without a single mention of the fact that most Linux distros now ship with Apache 2 and, yes, it has much better performance than Apache 1 because it can be run multithreaded.
This article should be scrapped and redone – shame on you, Logan!
Did David Adams got himself an Apple recently to write his articles on..?
The Opterons run great, and even outpace the Itanic system (which costs ~$20k compared to the $2k-$3k for the Opterons), but I agree with you. While this article was interesting, it would have been much better if they had used Apache 2 (preferably IN ADDITION to 1.x, but 2.0 only would be OK with me).
Still, looks like the Itanic is in DEEP trouble. Well, it was in DEEP trouble when it had no major competition for the x86 upgrade path. Now that it has competition, it seems to be searching for ice bergs to run into ground on and die.
Viva le Opteron!
On the downside, the power button is tiny and hard to press, and the keyboard and mouse ports were not labeled.
To which Mr. Rasmussen stated:
The guy is talking about a 1U Linux server that is usually used headless and powered 24/7.
<sarcasm>Very relevant, no?</sarcasm>
I’d say that was worth a comment. If I’m having to deal with keyboard/mouse ports or power buttons on 1U servers in my data center, I’M ALREADY REALLY REALLY STRESSED AND MY HAIR IS CATCHING FIRE AND WHICH STUPID PORT GOES WHERE AND WHAT THE HECK DOES IT TAKE TO GET THIS ACCURSED SYSTEM TO REGISTER THE FACT THAT I’VE PUNCHED ITS STUPID POWER BUTTON!
Additionally, it’s very human to use this kind of thing as a proxy. If the little things are right, then it’s more likely that the big things, less easy to see, are also right.
I’ve recently been looking fairly heavily into Opteron and Itanium systems. A web site I admin has a 32GB MySQL/Innodb database which with the 2GB memory limit of IA32 is starting to struggle. I’ve been hard pressed to find many Opteron systems that can handle 16GB of RAM and basically none which can hold more than that. Money is available for this but not an unlimited supply. A ~$4K server is fine and that leaves money to get a _lot_ of RAM. A 1U or 2U Opteron that can handle 32GB of RAM would be so nice… Ideas or pointers?
I don’t know what your time frame is, but if sutable motherboards aren’t out yet, I’d think they’d be out soon. That said, I’d call Dell or HP or IBM or someone and ask them what they can do for you in an Opteron that can hold that kind of memory. And have you considdered clustering two or more opteron machines? They are vastly cheaper than Intiums. Also, try contacting someone with AMD directly and ask them if they can help you.
hahah commodity servers on IA64? yeah as if. IA64 isnt commodity.
wendell dingus, a 32gb mysql db? buy a REAL db server, like an as/400. a 1u system with 32gb ram? do you know how much power that would draw? do you know how much space 32gb ram is? do you know the dimensions of a 1u server? nice troll!
32gb mysql db.. hahahahahahahaha
i think he meant 32mb.
— “a 1u system with 32gb ram? do you know how much power that would draw? do you know how much space 32gb ram is? do you know the dimensions of a 1u server?”
I don’t see how 8 slots with 4GB modules is so hard to believe. Are you saying that nobody makes a 1U with 8 slots?
64-bits is old-hat for Linux. Linux has been running on 64-bit machines for nearly a decade. Heck, it’s already running in full 64-bit mode on the G5 — even OS X (the G5’s native OS) isn’t there yet! This is news?
painfully so….ugh, Linux? Ready? It has been for a while! I remember years ago reading about how Compaq engineers benchmarked Linux (2.2 I think) against Tru64 on an Alpha and were shocked just how close Linux was in many of the tests (I believe within 30% much of the time, not bad considering). Didn’t the first Linux > 64-bit port happen in freaking 1994? Now, outside of that the article seemd decent – but the headline needs to go. It makes it sound as if 64-bit support is something new untested for Linux, it is not.
innodb_data_file_path = ibdata1:2000M;ibdata2:2000M;ibdata3:2000M;ibdata4:2000M;ibdata5:2000M; ibdata6:2000M;ibdata7:2000M;ibdata8:2000M;ibdata9:2000M;ibdata10:2000M ;ibdata11:2000M;ibdata12:2000M;ibdata13:2000M;ibdata14:2000M;ibdata15: 2000M;ibdata16:2000M;ibdata17:2000M
Actually 34GB allocated, about 32GB currently being used…
But anyway, 1U opteron systems that max at 16GB RAM are actually becoming fairly common and not ridiculously expensive. The only thing is that systems that can hold that much RAM using XEON 32-bit CPUs are also fairly common. The only real advantage to a 64-bit CPU is the larger memory address space. If you can’t get a _lot_ more memory in one, what is the use? That was my point…
Penguin Computing for instance:
http://64.240.166.228/store/altus-1000E.php
Power draw? Has a 350W power supply in it…
Anyway, 64-bit systems with lots of ram as database servers… natural fit. On the low end it will just be great when a reasonably priced box can be purchased that a lot of ram can be stuffed into. For my home machine 2GB is a lot.. For a server for this database I’d define it as >16GB.
>I don’t see how 8 slots with 4GB modules is so hard to >believe. Are you saying that nobody makes a 1U with 8 slots?
Well, if you go to pricewatch, you’ll note that the lowest price for a 2GB module is $1199. (4GB aren’t even available yet… when was the last time you actually did the specs for a real server?) Also, if you want to spend $4000 for a MySQL server…. run a smaller DB than 32GB. I have a friend who worked with the team supporting the main information Database at a major Midwestern University (Hint: It’s a Big 12 School) We are talking all the student and employee information, including class schedules, home addresses, grades, insurance data…..
It ran on a full blown DB2 installation on a REAL big-iron
IBM mainframe (no I’m not talking a Linux server on an old
PC I’m talking a full OS-390 badass mainframe).
Guess how big the entire DB was….. 3GB!!
No, I did not say 3 terrabytes, I mean 3 Gigabytes. And guess what? The amound of work that took place to keep it running was enourmous, because even though the size was relatively small, there could be upwards of 10,000 concurrent users at once, especially at class registration time. The admins had it tuned so tight that they could tell you exactly how many rows in a particular table were safe to lock at once without causing a crash.
The point is, your 32GB database is probably misconfigured. When was the last time you profiled queries to see which rows are locked most often? Are all of your data normalized properly? Can you archive data that is not queried? Are you using a database to keep files that should actually be kept in a normal filesystem (very common with big databases)? Your non-existant 32GB Opteron system for $4,000 will not solve your DB strucural problems. And a 1U server will not hold the necessary RAID array that any and all 32GB servers should be on to begin with…..
Well it’s amazing how you can analyze my database without even having a clue what it holds
It probably could be a bit more optimized than it is but that’s not the original point of this discussion. RAM is faster than disk and searching a database for something cached to RAM will be much faster. DB2 and big IBM mainframes are fine, for those who can afford them. I’m on a bit of a more “shoestring” budget than that. I’ve been evaluating things like linuxvirtualserver.org to loadbalance my database across multiple servers as well as other things. Another way would be to just have a _big_ server able to cache more in RAM and solve my speed problems that way. About the time I started this research Opteron systems were just beginning to become available from multiple sources. Comparing them to commonly available servers from SuperMicro the first thing I noticed was that you can’t put a lot more RAM in them than the Xeon systems. The advantage of a 64-bit CPU is mostly lost if you can’t use those 32(or 28) additional memory registers to ‘register lots more memory’.. That’s all.. Just giving a real-world example why this would be desirable.
PS. Row locking mostly has to do with _writing_ to a database.. I’m interested in screaming fast _query_ access to a relatively large database from something considerably less than 10,000 concurrent users. Mundo RAM is one way to accomplish this, but only one…
PSS. 32GB of database data.. Files and other data not appropriate to store _in_ the database are stored in the filesystem with database references to them. 2TB currently and soon to grow to >3TB.. Yeah, TB as in terabyte. There are real-world _large_ systems run on Linux of which the details commonly can’t be made all that public.
Errm.. when I was young, it used to be a FEATURE that the power button is difficult to press or even hidden away so that not every unconscious dumb ass would switch it off by mistake. When I saw that the very first time, I thought to myself: How st00pid a person does it take to switch a system off where you actually don’t want to..?!” — Well, shit happens and appearently I turned out to be one one these people. So, if a system is meant to run 24/7, I better press the power twice or harder instead of shutting a 24/7-system down hard where I didn’t want actually to. Get over it.
So, if a system is meant to run 24/7, I better press the power twice or harder instead of shutting a 24/7-system down hard where I didn’t want actually to.
Agreed. In fact many boxes that must run 24/7 don´t even _have_ a power button.
I have far too much time on my hands… sigh.
Well, if you go to pricewatch, you’ll note that the lowest price for a 2GB module is $1199. (4GB aren’t even available yet… when was the last time you actually did the specs for a real server?)
Uhm. If you run a business and want to buy a $20,000+ machine, you don’t typically go to pricewatch to price individual components out. You go to a VAR or to your rep at a reputable vendor who has at least 8×5 onsite support so that if the damn thing blows up you can get it replaced that same day… not to mention not having to wait a week to do a full burn in on all the components.
Sure, it’s nice to know how much of a mark up you’re paying, but at the end of the day IT costs dwarf hardware costs.
As it stands, 4 GB DIMM modules are available from Samsung based on Micron’s 1 Gb DDR chip, but won’t be in volume until later this year, so good luck finding it on pricewatch.
“Didn’t the first Linux > 64-bit port happen in freaking 1994?”
Yeah – Intel delivered one of the first test Itanium boxes to Linus’s office where he had a 64 bit Linux system running in 48 hours according to Bob Young’s book
Actually, the first 64-bit port of Linux was the Alpha port in 1995. It was mostly working by March, and by May (with release 1.2.5) it was self-hosting. You can find a history here: http://www.rusling.uklinux.net/linux/alpha/alpha-history.html
Really, get an old alphaserver from ebay or something for a couple hundred bucks … You get true 64bit arch, multiple pci busses, all scsi box. Even a today’s PC with similiar configuration is more expensive.
You have many distros to chose from, redhat, suse, debian, gento …
Latest alphaserver still outpreforms xeon boxes in floating point, even if xeons run twice the frequency of alphas. Go figure. Clock for clock, Alpha is still the fastest overall CPU today.
Hint: Athlons actually use the ev6 bus of alphas