Long answer: IBM's VM operating system can nest itself to nearly any reasonable depth, and they don't even have to be the same version. Inside the innermost level of that, you could nest a Linux instance.
If you really want to go wild, but don't care if it's practical or not, there's a program that emulates an Intel PC under Linux, and it has been ported to S/390 Linux. You can then run Hercules (an S/390 emulator) within that and be back where you started but with huge overhead. I said it's not practical. I'm pretty sure it's been done, though I forget the exact configuration.
I used to work with a mainframe admin who routinely ran a complete OS/390 (formerly MVS) environment underneath VM, for testing OS/390 patches.
By the way, I've also been a beta tester of the system. The account is a full virtual hardware environment, with root access to your virtual machine. The whole thing runs under IBM's VM operating system which acts as a hypervisor like VMWare(tm) but with extremely low overhead because the S/390 CPU has better virtualization support in hardware than Intel does.
They are using SSH with hostkey authentication in addition to the login/password authentication. This is so that they can safely allow a direct root login remotely. You can (and will, if you're smart) also use IPCHAINS to create a personalized firewall on your virtual machine. Isolation from vm to vm is extremely good because of the hardware-assisted virtualization.
IBM would be the first to admit that the raw MIPS numbers of a mainframe, even a big one, don't compare to a huge cluster of RISC or Itanium boxes. The forté of a mainframe is I/O bandwidth and security, and they are marketing it as such. Big companies are finding that running Linux on the S/390 concurrently with their traditional mainframe applications allows a best-of-both-worlds integration that is otherwise very cumbersome and difficult. By the way, there is an internal bus-speed network within the mainframe chassis that carries TCP/IP traffic from vm to vm at gigaBYTE per second speeds. So if you have a few dozen virtual Linux servers communicating with one another and with OS/390 or VM/ESA mainframe apps via IP, network bottlenecks are seldom a problem.
IBM's engineers have been very responsive in fixing bugs and listening to suggestions, and they've set up a discussion e-mail list with several of their engineers lurking and posting. This is a very new thing for IBM, and there will inevitably be glitches, but give it half a chance. As for the coolness factor, that's certainly there but it's not what IBM is pushing. It's not fair to judge this offering based on whether it does a great job of running seti@home or Quake. Judge it on whether it's useful as a porting lab for Open Source applications and for IBM to learn about hosting huge numbers (tens of thousands) of simultaneous Linux instances. They have kernel gurus working on performance issues and are contributing what they learn back to the kernel team.
Incidentally, for those who have the IBM == big_company == bad_motives mentality, you might be surprised to learn just how many internal hoops had to be jumped to make this happen. Those hoops were jumped by a team of mainframe geeks who think Linux is way cool. Companies are made of real people, and they aren't all selfish jackasses just because they work for a big company. Does IBM hope to sell iron? You bet. Just like any other business. And while they're doing it in this particular way, they are causing some enormous customers to quit ignoring Linux and take us all seriously. How is that a bad thing?
Note: This post is my personal opinion, and I don't speak for IBM or for my current employer.
You are correct that ECN is not enabled by default in the stock kernel. I neglected to mention that in the article, and probably should have. The article was written from a tutorial standpoint, using the example problem to illustrate the issue with routers not supporting ECN.
As several folks have pointed out, reading the kernel help does provide insight into the problem. Since so many web sites do work just fine with ECN enabled, though, it is very easy to forget when you find an unreachable site that several days ago you turned on this experimental kernel feature just to see what it did. This was my situation, and from other posts I've seen here and e-mails I've received directly, it appears that others have encountered the same situation. Hence the article, to help others avoid frustration.
Slashdot Mirror
Long answer: IBM's VM operating system can nest itself to nearly any reasonable depth, and they don't even have to be the same version. Inside the innermost level of that, you could nest a Linux instance.
If you really want to go wild, but don't care if it's practical or not, there's a program that emulates an Intel PC under Linux, and it has been ported to S/390 Linux. You can then run Hercules (an S/390 emulator) within that and be back where you started but with huge overhead. I said it's not practical. I'm pretty sure it's been done, though I forget the exact configuration.
I used to work with a mainframe admin who routinely ran a complete OS/390 (formerly MVS) environment underneath VM, for testing OS/390 patches.
By the way, I've also been a beta tester of the system. The account is a full virtual hardware environment, with root access to your virtual machine. The whole thing runs under IBM's VM operating system which acts as a hypervisor like VMWare(tm) but with extremely low overhead because the S/390 CPU has better virtualization support in hardware than Intel does.
They are using SSH with hostkey authentication in addition to the login/password authentication. This is so that they can safely allow a direct root login remotely. You can (and will, if you're smart) also use IPCHAINS to create a personalized firewall on your virtual machine. Isolation from vm to vm is extremely good because of the hardware-assisted virtualization.
IBM would be the first to admit that the raw MIPS numbers of a mainframe, even a big one, don't compare to a huge cluster of RISC or Itanium boxes. The forté of a mainframe is I/O bandwidth and security, and they are marketing it as such. Big companies are finding that running Linux on the S/390 concurrently with their traditional mainframe applications allows a best-of-both-worlds integration that is otherwise very cumbersome and difficult. By the way, there is an internal bus-speed network within the mainframe chassis that carries TCP/IP traffic from vm to vm at gigaBYTE per second speeds. So if you have a few dozen virtual Linux servers communicating with one another and with OS/390 or VM/ESA mainframe apps via IP, network bottlenecks are seldom a problem.
IBM's engineers have been very responsive in fixing bugs and listening to suggestions, and they've set up a discussion e-mail list with several of their engineers lurking and posting. This is a very new thing for IBM, and there will inevitably be glitches, but give it half a chance. As for the coolness factor, that's certainly there but it's not what IBM is pushing. It's not fair to judge this offering based on whether it does a great job of running seti@home or Quake. Judge it on whether it's useful as a porting lab for Open Source applications and for IBM to learn about hosting huge numbers (tens of thousands) of simultaneous Linux instances. They have kernel gurus working on performance issues and are contributing what they learn back to the kernel team.
Incidentally, for those who have the IBM == big_company == bad_motives mentality, you might be surprised to learn just how many internal hoops had to be jumped to make this happen. Those hoops were jumped by a team of mainframe geeks who think Linux is way cool. Companies are made of real people, and they aren't all selfish jackasses just because they work for a big company. Does IBM hope to sell iron? You bet. Just like any other business. And while they're doing it in this particular way, they are causing some enormous customers to quit ignoring Linux and take us all seriously. How is that a bad thing?
Note: This post is my personal opinion, and I don't speak for IBM or for my current employer.
As several folks have pointed out, reading the kernel help does provide insight into the problem. Since so many web sites do work just fine with ECN enabled, though, it is very easy to forget when you find an unreachable site that several days ago you turned on this experimental kernel feature just to see what it did. This was my situation, and from other posts I've seen here and e-mails I've received directly, it appears that others have encountered the same situation. Hence the article, to help others avoid frustration.
No insult to the kernel team was intended.