Journal eno2001's Journal: The Madness of Me 6
In my quest for a machine that almost never stops, I've come across QEMU , User Mode Linux, OpenMosix and Xen. So far, I've only played around with QEMU. My next target is Xen. From what I can see QEMU is neat, but the performance hit is too great for my needs. Just to give you some idea of what I did and what I noticed:
1. Compiled and installed QEMU
2. Downloaded a few prepackaged freeOS images and ran them
3. Focused on Mandrake 10 specifically to see how the desktop experience would be...
Number three here is the key. I ran QEMU on a P4 1.8GHz machine with 256 Megs of RAM. The Mandrake 10 image took quite a while to boot up and was very slow. Slow enough to really be unusable other than as a curiosity. I think QEMU might be useful for running some DOS apps that you absolutely can't live without, but it's not really good for day to day use as a desktop. Of course, that's not what QEMu was made for either, but I just had to try it for myself.
I was eyeing OpenMosix and I probably will still try it in the future for another application, but it's starting to look like the Xen project is more along the lines of what I want. The "work" involved in getting it to run is in the porting of your guest OS to the paravirtualization platform Xen. Xen is a virtual machine monitor kind of like VMWare, but it's much more efficient because it REQUIRES that your guest OS is tuned to it. The good news is that the ABI is not affected at all so you don't need to recompile your x86 binaries. All that really needs modification is the kernel of your guest OS. It is also implied that Xen can cluster hardware beneath it to support the guest OSes. Some of the really cool stuff seems to be that you can actually move a live VM from one node to another in your Xen cluster without the user noticing. You can also suspend and restore the VMs much like VMWare so that your VM is never aware that it ever went down. I still need to find out if Xen can run on top of Mosix as crazy as that might sound.
My reason for writing this is that I was wondering if anyone out there has experience with User Mode Linux and if you would care to share experiences with me regarding UML. It looks a little less likely that UML will work for what I want to do, but I'm keeping my mind open to it.
In related news, I think I am going to go AMD for my new monster machine. The only questions left in my head now are:
1. Should I go SMP with dual Opterons?
2. Should I go cluster with multiple individual Athlons?
3. Which will have the best price vs. performance ratio?
Right now, I'm very tempted by the idea of clustering a few Athlon 64s for the raw horsepower. This option also appeals because I can build the cluster slowly over time. I can start out with just two Athlon 64s and add more later. With SMP using Opterons, I'm going to have to fork out all the cash in the beginning.
I did this with my current app server back in 1997 (Dual PII 450 with 768 Megs of RAM). At the time, the system cost me about $2800. But the benefit is that I haven't had to buy a system in about eight years and it still keeps pace with my P4 2 GHz system for most basic uses (web browsing, e-mail, wordprocessing, GIMP, etc...). But, clustering changes everything. I can buy cheaper boxes and spend even less money now and get something similar to SMP (not parallel) and grow the system over time. I don't think that it should be necessary to buy a new machine to use new software every two to four years. So far GNU/Linux has proven that to me with my current SMP box. The money should all go to the hardware, and the effort (on my part in terms of configuration, and set up) should go into the software. My goal with this new system is to have a system that I can use for the next 8-10 years and still be running current software much as I've already done with the dual P II. Barring any major shifts in hardware platforms, global disaster, alien invasion, etc... (Remember! Planet X is coming and the Niburu will soon return!
Hmmm... Upon further reflection: I have multiple goals it would appear:
1. To have a computing system that will last nearly a decade
2. To have maximum performance for the lowest price
3. To have a machine that is "always on"
4. To be able to create new virtual machines as needed and dispose of them just as easily in one central location
5. For the system to be able to present a few desktop VMs as well as server only (web, SQL, mail, VoIP, etc...) VMs
6. Efficient power consumption (that's the "killer app" for the AMDs from what I've read/heard at the moment)
My god, I think I'm becoming a "mad scientist"... If I ever get this thing off the ground and it works as well as I'm hoping, I can just picture myself in my lab yelling "Alive! It's ALIVE!!!"
Hmm (Score:2)
As far as the "always up" machine, I want to get something like this for home too. I will probably just go the route of regular machine+UPS+basic raid card (3ware raid cards rock on fedora!) As far as one self-contained machine, there really isn't any way to ma
Oh one thing (Score:2)
I forgot to mention, 256MB of ram seems really low. I bet if you upped the ram the speed situation would improve quite a bit.
Yet another one: bochs (Score:2)
I couldn't get QEMU to work at all, but bochs worked great (well at least for booting dos inside linux). So far it looks slow, but the machine I am testing it on is slow as hell too. Anyway just wanted to add that to your list if you haven't seen it already.
Have you tried BOCHS? (Score:1)
My two cents ... (Score:1)
- need (want?) a powerhouse to experiment with one/some/all of: SMT, NUMA-aware threading, clustering, and/or high reliablity.
- want to experience/help the jump to 64-bitness on the cheap BUT not sacrifice the longish term (5 years?) investment with underpowered hardware, i.e., discounting old Sun hardware (for example) right away
My own jumping off point was also similar: dual PIII Xe
yeah (Score:1)