Posts by Bronek Kozicki 2859 publicly visible posts • joined 6 Sep 2007
Let's face it, software is now a necessary part of daily life and as such should be freely available to anyone.
It would be nice, if it was true. I am happy to support software I've written and get paid for that, but if everyone wanted to just use my software for free, I would be forced to find another profession.
Agreed.
My motherboard, SuperMicro X9DA7 (two sockets LGA2011 for E5-26xx) can do mirroring too. It's not new feature.
Extra bandwidth in connection to mirroring (so apparently there is no performance penalty compared to current generation) seems to be new, to me.
The article left me confused. Memory mirroring with lockstep is not a new technology.
It would seem that actual news here is double the bandwidth, which can be utilised in one of two ways (none of which is new). Either you take the lockstep, meaning pair of DIMMs handling shared duties (mirror each other) and only half of the installed memory size is available to OS, or you take those as independent modules, exposing full size of your DIMMs to the OS.
Either way you will use ECC memory, so the only "unreliability" implied here would be when you actually have unrecoverable memory errors, i.e. more than 1 bit flipped. In which case lockstep would presumably allow you to continue work, hopefully with your data intact (thanks to mirroring performed by CPU) while performance mode will force you to replace memory modules (taking the machine down, first). Either way, you would see it coming since failing ECC memory modules would first flip single bits (recoverable and logged if you have decent OS), so you should be able to replace the memory during e.g. weekly machine bounce, before unrecoverable errors happen. There is nothing new here. Except for the bandwidth, it would seem to me.
Now, this is really interesting, I just found out about FreeBSD bhyve , and on its FAQ page found this gem:
Q: Does bhyve support VT-d PCI device pass-through?
A: Yes, on Intel CPUs. See wiki.freebsd.org/bhyve/pci_passthru
Q: Could bhyve support UEFI/BIOS?
A: Yes, this is a priority because it would simplify the booting of non-FreeBSD OSs and implementing integrated video support.
Q: Could bhyve support VGA graphics?
A: Yes, this is closely related to UEFI/BIOS support.
I understand that VGA passthrough is also being actively worked on. It is not ready yet, but I will want to try it when enough progress has been made :)
If I were lone wolf then I would just put a server anywhere. As the things are, I must consider my wife and children. And yes, moving out is definitely part of a plan, but if you haven't noticed property market is behaving rather strangely, especially in London. So, this will take some more time and preparation. Printer is on a stack of drawers which sits on top of a desk next to 30" monitor, under which is large document shredder, subwoofer, my legs, spare toners and lots of cables. YES IT IS F*G CROWDED HERE. You have to come and see what builders call "flat" in this part of the world.
So, the hardware is ordered.
All should work well with VGA passthrough and even if virtualisation turns out to be too difficult, it's going to be some nice Linux server (and a GPU for my old PC)
* SuperMicro X9DA7
* 2x Xeon E5-2630V2
* Kingston DDR3 PC-1600, Registered ECC
* Sapphire Tri-X R9 290X
I guess, if I really have to make this Linux a separate server; I can always put in into 2U rack case and slide under a bed ;D
I think you captured my requirements almost perfectly and thank you for this summary. There is one more I assumed is implied (from subject, perhaps?). I do not quite believe that "virtualisation isn't there yet" unless my own experience tells me so. So, the implied requirement is : make the virtualisation do the work. I will share here later if it worked (or it didn't).
PS Really, one average width tower case is as much as I can fit in here. And perhaps some tiny NAS or microATX in the corner (but this goes against Linux having CPU power I want it to have).
"Building gcc very often" is a batch job
For me, making a change in a complex program with very little type safety in its design is a continuous process of making changes, building it and then running tests. And waiting, while build and tests are running.
If you have better workflow for making gcc changes, I'd love to learn about it.
The size is the problem here. I want "non compromise" solution where Linux has lots of power (two socket motherboard size E-ATX) for serious stuff like building gcc very often. For Windows I also want "non compromise" solution with strong (and large by necessity) GPU card and some interesting peripherals. This would normally require two large boxes, for which I do not have space. But, I do not need much processing power for Windows and I do not need any interesting peripherals, not even a GPU, for Linux. Ideally Linux will run headless and will provide network services for Windows guest (mostly sshd and filesystem).
These system are meant to complement each other, so why not use the parts in the same way and actually put them in one box? If this works, I will also gain something no two-box solution can do: flexibility to move resources between systems as I see fit, simply by configuration tweaks. And there is convincing evidence that such systems have been built and are known to work, also in this thread.
Gordan, many thanks! I checked the article on modding Nvidia cards then checked the Quaddro prices, and my eyes almost popped out. Yes I would pay this sort of money for set of CPUs, lots of memory and server motherboard, but for GPU not faster than 780ti ... greedy b*rds!
I am not sure I want to buy their hardware, but if AMD is so much harder to get to restart properly without physical reboot .... :-/
Thank you for this. I've had some exposure to FreeBSD (or was it OpenBSD?) at the time when BSDi meant something, so it is not exactly a new operating system to me ;)
I know ZFS and FreeBSD fit like glove to hand, but I cannot see any virtual machine which would work on FreeBSD and allow for PCI passthrough for Windows guest. This feature is absolutely crucial for me; FWIW I actually plan to run the host OS headless.
I have MegaRAID 9265-8i and can buy some extenders to it - that gives me ridiculous amount of storage (and the single case I have here is large enough for at least 8 HDDs). So this is not going to be a problem, probably.
You gave me good reason to consider ZFS again, thanks for that. Perhaps with enough RAM given to host it will work fine.
@AC
no space for two computers here, really. I could put second microATX sized machine somewhere in the corner but it would be constantly at the risk of being tripped over, disconnected from network or power, or damaged in other ways. Even if place is found away from people (I do not see how - this is London flat, aka "rabbit hutch") it would simply not have enough CPU power for things I want to do under Linux. Headless host and guest Windows taking over fair bit of peripherals (but not much of CPU/network/disk/memory) with help of PCI passthrough is what I'm aiming for.
If the host is running headless and Windows is given PCI passthrough to GPU (and some USB controllers), I do not quite see this much potential for problems. Yes a lot can go wrong, but not as much as you seem to think.
On the other hand keeping one box on top of another, when the children are fidgeting just next to this stack of boxes, or even better trying to sit on it .... that sounds like a lots of fun ;) I simply do not have the space for two boxes and dual boot is also out of the question, period.
@AC thanks, I will investigate JEOS. It definitely seems like promising direction. Name "Oracle" is little off-putting, though (but hey, it is still FOSS so I'm not going to complain).
And yes, I am definitely not going to mess with Linux hypervisor, any actual work or experimentation is going to be in a "sibling" virtual machine. This hypervisor is actually the only Linux OS I want to have an option for support.
@John Doe 6
I do not quite see where did you get the spy aspect from. Serious use can be at home, that's what my children call "work" when I'm doing such things as trying to write nice but tricky lockfree data structure or write another gcc patch. Both tasks require extremely robust testing with access to as many cores as possible, later also access to all the disk IO I can get. Both are best done under Linux. I understand it is arguable how serious use it is, but when one gets involved into C++ standardization in his private time, then the lines get blurred.
Since I only have space for one good PC at home and my family needs to access it too (I have a family here!), I want to make it suitable for the tasks they usually need it for (which all run under Windows), but without sacrificing the power I want for Linux based tasks.
With PCI passthrough I believe I can play 3D games quite easily and with native performance. Let me explain - Windows is only meant to be nice façade, but with full access to few devices including graphics card and USB controllers. The fact that PCI passthrough is only available on XEN and KVM is exactly the reason the topic is set as it is.
Underneath this façade I intend to do other more demanding tasks, which will require heavy IO and CPU - and which make best sense to run under Linux. But also which can be headless, i.e. via ssh session.
I am very familiar with double boot and it is not suitable here - the family might want access to PC while I'm in the middle of something I simply do not want to interrupt. With the desired setup I can simply migrate my tmux session doing the work over to ssh on my underpowered laptop, while allowing them to use Windows as-if no-one else was using the computer at the time (cheating, but useful). Or even mess with Windows without stopping that other task. I simply have no space for other box nor the desire to put the kind of processing power I'm after into a laptop. It would be very expensive, very heavy and not practical at all.
Oh and one more thing. I do not actually play games much (my sons do, but wife and me control both time and content). It is not a request to help me put "gaming rig" in a VM.
@Gordan
many, many thanks - your post was the most helpful in the whole thread so far. It is very likely I will go with Xen. However, since I'm not in a hurry, I will spend some more time investigating KVM as well. Apparently there are some success stories with Radeons on PCI passthrough, even though it is fiddly and not really supported. PCI hot reset feature added to kernel 3.12 is promising, too.
Also thank you for a link to xen.crc.id.au - this is exactly the kind of resources I was looking for!
Thanks h4rm0ny; I've been running Linux under VMware and I'm unconvinced by this:
Clearly if the OP wants to do heavy compiling, then all you would need to do would be to install Windows on that set up, tell VirtualBox to give the guest (Linux) 14 cores and 28GB RAM and you're going to have a mighty powerful compiling machine.
If Windows kernel is given all the cores and memory, there is no telling how much it will actually consume, at the expense of Linux guest. This is not Windows fault, it is simply how some programs (games) behave. But if Windows is host, it simply has to be given all. I want Linux to have the primary access to resources, not the other way around.
You are right that I will most likely give more resources to Linux than to Windows - I've given more thought to it and this seems inevitable. But again, running Windows under VM gives me an option to simply tweak its configuration and restart Windows with access to extra 2 cores if something a bit more demanding needs to be done.
Thanks for all the advices so far, even those "run Windows natively". Let me explain why I do not want to do that: I've been running Windows natively "always" but since I started using Linux, I've grown to respect its kernel. I personally believe Linux kernel is more robust and more economical than Windows one, esp. the scheduler and memory managment. I've been running Linux as guest on Windows both in VirtualBox and in (paid) VMware, and while the experience is pretty good for less demanding tasks, it is awful for full gcc build. Especially if one is actually hacking gcc, and thus has to run the build again and again - it takes an awful lot of disk IO and CPU power.
What I want to do is to let the thrifty kernel own all the resources since it's less likely to waste them, and more likely to let me utilize all the cores, memory and IO I need to run full gcc build, even from within guest Linux. It does not bother me if it's headless, in fact that's how I like Linux best. The less economical Windows is to become a kind-of front end, started automatically, with full access to keyboard, GPU, USB devices etc., but in its VM configuration I decide how much memory and cores it will be allowed to use. The plan is to setup a Xeon machine with some 10 cores, of which 4 (number which I will tweak as needed) is given to Windows, and 4 or more to sibling Linux VM and rest to host for tasks such as SAMBA.
Primary use of this is to let the family use Windows as usual (games, internet etc), while I build gcc in the bacground, having started the task over ssh from my laptop. Or while I do anything else either via ssh console or X session, from my underpowered laptop. Since games can potentially eat up all the CPU allocated to Windows, I really, really do not want Windows to be host for this. It makes much more sense the other way around. In a word, Windows is primary use only if all you can see is front-end. There will be much, much more going on behind the scenes and managed through ssh.
The time is slowly approaching for me to rebuild my home PC. Since I've been using Linux at work for the past few years (ssh only, and happy with it), I want my new machine to be running Linux as a host system, and Windows on top of it. Not full migration because I've invested good money into software which is Windows only, also want to play games (which aren't ported to Linux yet). Also I find Windows GUI more ... attractive than Linux windows managers.
Pretty standard so far. What's different:
* I want to use PCI passthrough for AMD GPU; possibly also for LSI MegaRAID (unsure about this)
* I want to be able to give exclusive access to selected USB ports to Windows (to use devices I do not need Linux to know about)
* I want Windows to start automatically "as if" it was regular Windows-only machine, with the host OS not much different (for layman - i.e. family) from BIOS. In particular: no manual interaction needed to start Windows full screen with access to all the hardware I've given it.
* I will be using Linux from Windows guest over virtual network almost exclusively, i.e. ssh and possibly X sessions (X-server being Cygwin)
* I'm fine to put good money into hardware, 32GB RAM and 2x Xeon are not out of question. Also happy to split resources in half - Windows does not need to have all the goodies.
* I will be using Linux for some performance sensitive stuff, although nothing critical. Almost always related to programming in C++ (I wrote some patch for gcc in the past, want to do more of this, and gcc build takes lots of resources).
* I'd like to be able to buy support for my Linux installation, so distros such as RHEL or OpenSUSE are in
* I'd like to be able to use btrfs with confidence (strong point for OpenSUSE but I can wait few months for official support in official release of RHEL7). Specifically, I want self healing on RAID with checksums.
* I'd like to share filesystems between Linux host and Windows guest - if there are options better than SAMBA4 I will be happy to learn.
* I do not want to run Linux full screen GUI. I dislike windows managers in general, and also prefer tools like ssh/tmux/vim for work; on the other hand running Linux full screen in text mode is just waste of space and I do not need it. Putty ssh is best tool for me, for actual physical screen I can put Linux in picture-in-picture, second input on main monitor (as long as its text only).
* I do not want second keyboard, no space on desk for it, but how will switching of focus work? My keyboard has no special functions whatsever (but I love it because it's mechanical)
* This is not for playing with or learning VMs (which is why I didn't add this to the other topic), this is for serious use where VM is supposed just to stay in background and let me do my work. I have some experience with VMware (and old license I could upgrade - but they do not support PCI passthrough last time I checked).
Question: which combination of distro + virtual machine would work best for me? Thanks!
For one thing, sitting in front of a large screen in order just to read something (largish email, article, whatever) is not my preference - I'd rather print it, put my legs up and read away from the computer. Sitting in front of a computer is what I do at work, doing the same at home seems like too much work.
unless you live under a rock, it should be plenty obvious that the current pricing model of cheap ink printers actively discourages from actually using them to print anything. These devices are meant as a office decoration, not as printers!
If you want to actually print anything, you need to perform due diligence on the cost of refill and this will most likely direct you to either more expensive models (e.g. some Epsons) or to laser printers.
Both the rootkit affair and PS3 other OS taught consumer not to trust Sony.
No matter whether you are consumer, corporate executive, or financial analyst, you must understand that losing consumer confidence is a terrible blow for the company. Sony never properly apologized for both blunders and thus the impression is that they never learn, in which case regaining consumer confidence is going to be difficult, to say at least.
If this sentiment backed by low sales figures then it really does not matter what you think about such blunders. Even if these were perfectly legitimate marketing strategies, this is irrelevant. You have a company incapable of learning and thus unable to enter new markets, and also unable to sustain its existing markets. Thus, junk company.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
