Slashdot Mirror
Linux SMP Round-Up
Dual Minds writes "LinuxHardware.org is at it again and this time they cover three of the finest boards on the market. This review covers three dual processor Xeon boards and they are the only site that ever does Linux reviews (at least on a regular basis). Here's a peak: "First thing is that all E7505-based boards are basically the same on the surface due to the basic features of the chipset. They all have dual processor support, support for dual channel DDR, and support for PCI-X up to 133MHz (to name a few). Once a manufacturer gets their hands on the board though, features can be added or it can simply be left as is." Very in depth and some sweet hardware."
I'll go read the article now.
Old people fall. Young people spring. Rich people summer and winter.
I would like to see a comparison of Linux 2.4, Linux 2.5, FreeBSD 4.8, and FreeBSD 5.0 on the same hardware. FreeBSD fanatics like to toot their horns, but where are the benchmark results?
btw, LinuxHardware.org is nearly slashdotted, so their Linux server knowledge must not be so great after all..
cpeterso
An actual comment on the story...
When reading through the review, I noticed that they only list standard benchmarks, and then a kernal compile benchmark. They never list the actual distribution of Linux used for testing the system. In my experience, the actual performance of a system is dependant on that. I know I had a system that just dragged running Mandrake, but loved Debian to no end. I'm not sure if it's just the kernal base of the system, but most of the actual distributions have some sort of performance optimization (I think) for the overall system performance. I mean, kernal complilation time is great, but what I'm more curious about is the day-to-day operation.
I guess I've just read too many reviews over the years that focused on benchmark numbers and didn't give any information about performance under everyday use. If this is something geared for Linux, I'd be more curious about numbers like Networking performance, data-access numbers and things like that.
My other curious question is how accurately does UT2k3 and Quake 3 show the power of a Dual Processor Xeon system? Quake 3 supports MP systems, but it has never been shown to make much difference except on large server environments. They give us video-benchmarks, and for Quake in particular, there's a limit that was hit long before these processors and chipsets that was somewhere next to overkill.
I guess I'm just being nit-picky, but I think a Linux Review for a system should concentrate on strengths, and not benchmarks that would be similar on a Windows system made to run games.
Before this, someone pointed me to Dell's Finacial Services' page of good deals (and no OS tax!) on lease-return laptops. After that, a friend of mine called to tell me that a Ford dealership nearby is selling a 2002 convertible Mustang GT for below invoice with 0% financing over 4 years. And don't get me started on what I could do with a Fry's or a Best Buy right now... Oh, the agony of being such a consumer whore...
It'll be a kick-ass ring, though. I highly recommend browsing this thread before making decisions on engagement rings -- good info even if, like me, you want to go with a diamond regardless of the fact that you're getting ripped off.
(posting anonymously to avoid my girlfriend seeing this post a la Murphy's Law).
If you need hardware like this, then you need Support. That's what attracts people to Sun (and now Dell, for instance). And if you need support, you'll take whatever board your System Integrator uses in their boxes.
To wit:
If you need this, you'll buy it from someone.
If you buy it from someone, you have no choice of HW.
Thus, this review is useless.
I want to delete my account but Slashdot doesn't allow it.
4 cpus for the price of 2? No.. that's not what hyperthreading is about.
At least, not from what I've gleaned from all the documentation out there.
Hyperthreading is about optimizing the pipelining features of the processor... wheras normally. If the processor knows that 2 instructions are independent of each other, it can run whatever stages of them it has roon for in the pipeline, concurrently. Normaly, preduction and whatnot have to be done, and this is only somewhat effective.
By forcing the OS to treat ti as 2 processors, it now has a clue as to which instructions are definately unrelated, as the higher layer OS has already decided they go to separate processors.
So Hyperthreading is really using 2 virtual processors to better use up the resources of a single processor.. so for some operations it may yield near double the perforamnce, but overall, there is no way this is going to give you the same boost as the equivalent number of processors will.
Yes, linux currently supports hyperthreading. You will see that 4 processors show up on a dual processor xeon system.
Linux does support hyperthreading. 2.4.20 recognizes four processors on my dual Xeon servers, without any tweaks. I think it's pretty nice -- I'd say there's between a 5% and 25% pickup in performance, depending on what you're using it for (generic vs. optimized integer code).
According to a geek.com article, Linux was actually the first operating system to officially support hyperthreading, and that was in late 2001.
The PCI-X controller used in almost all of the E750x workstation/server boards is really expensive and adds to the complexity of the board layout and design. It seems that Tyan decided to forgo that chip in order to keep the cost of the board down while making up for it by adding Serial ATA (but no FireWire like it's larger Thunder i7505 brother).
One board that I would like to have seen reviewed is the Supermicro X5DAL (with or without Serial ATA RAID) as it does include PCI-X slots, but it is also a standard ATX-sized motherboard. It only has four memory slots, so that may have changed some of the memory timings and possibly have improved some of the scores by a small amount.
One a side note, FreeBSD 4.8-RELEASE users will also benefit from the newly added support for HyperThreading found in all P4-based Xeons and the 3.06GHz P4. More info can be had here. I'm not sure if that feature is also available in 5.0-CURRENT (I would think it would be MFC).