Originally posted by kebabbert
View Post
Announcement
Collapse
No announcement yet.
Oracle Plans To Bring DTrace To Linux
Collapse
X
-
Originally posted by kebabbert View PostCool. But there are also people that thinks that C# is NOT a clone of Java. As everyone knows, this is wrong. MS released a Java implementation called J++, But J++ and Java was not really compatible, there were some slight changes in J++. This allowed you to write programs that could not be run on Java, the famous MS strategy of "embrace extend and extinguish". Sun sued MS because MS is fragmenting Java. Sun won so MS stopped releasing J++. Some year later, MS released a new cool language running on a Virtual Machine - even though C# only runs on Windows. Back then, there was no point of running software on a VM, if you tied it to only one OS - because VMs were slow back then. And early version of C# were nearly identical to Java. But, I have talked to developers that dont think that C# is a Java clone. So probably there are people that dont know that BTRFS is a ZFS wanna be and heavily inspired by ZFS.
No, I dont think so. I have not claimed that. IBM did lot of virtualization tech, back in 1970 or earlier. But Solaris pioneered Containers, which IBM has copied. And now Linux is copying too.
Virtuozzo used the the containers that solaris later implemented back to at least 2001 (of course, IBM had similar tech decades earlier but that was seriously locked up and optimized to their systems -- my understanding is that we are still trying to catch up to IBM when it comes to virtualization but I have no idea in what particular areas they are still superior).
OpenVZ dates to 2005 (right around the same time as Solaris Containers were released, IIRC.
See this http://wiki.centos.org/HowTos/Virtualization/OpenVZ.
Apparently linux-vserver is quite a bit older (http://linux-vserver.org/Overview).
Again, this is all old stuff to IBM. That's what you get when you've been around since, basically, the beginning of digital computers and spend billions every year on research (a lot of it is simply fundamental, i.e. basic science, research which I have no idea how they justify but then again I recall reading that Microsoft did similar fundamental research at one point).
Here's one more general overview of things:
Comment
-
Originally posted by liam View Posti don't want to get involved with this bickering beyond this point.
(of course, IBM had similar tech decades earlier but that was seriously locked up and optimized to their systems -- my understanding is that we are still trying to catch up to IBM when it comes to virtualization but I have no idea in what particular areas they are still superior).
...
Again, this is all old stuff to IBM....That's what you get when you've been around since, basically, the beginning of digital computers and spend billions every year on research
Regarding Containers, they are new to IBM, just as I said. IBM is calling it WPAR, and IBM implemented Containers in AIX v6.1 which came in year 2008. The IBM Container implementation is the latest of all OSes, and way many years after Solaris and Linux. So in a sense, IBM is copying others. Just as IBM Probevue is a copy of DTrace.
From 2007:
"I can say I am impressed by both the upcoming POWER6 processor and the new AIX 6 release preview . They both are innovative, and come with a large set of new or updated features. For the purpose of this entry, I will concentrate on a little side-by-side comparison between the features announced for AIX, and those available in Solaris.
The Workload Partitions (WPAR) on AIX seems to be very similar to the Containers, or Zones, which is an Virtualization feature found on Solaris."
In year 2008, IBM says that WPARs is a new innovation, from the IBM web site (I did not know that IBM innovated Containers? Calling it "bold new innovation"?).
"WPARs are a bold new innovation, implemented within AIX 6."
Virtuozzo used the the containers that solaris later implemented back to at least 2001
OpenVZ dates to 2005 (right around the same time as Solaris Containers were released, IIRC.
But the Solaris Container concept is much older, but run under different names. Earlier, Solaris had something called BrandZ which is the same thing. And before that, Solaris had a similar technique where Solaris tried to run Linux apps in a container, called Lxrun. But Sun found out that implementation is not good enough so they reworked it once more, to improve it. For instance, Solaris Containers allow to install other OSes, which you can not do in Linux implementation. So when you look at the earlier attempts at Sun to do a good implementation of Containers, you need to go back to 1999, which is before Linux version. And as we all know, Linux looks at other OSes and implement their own version, but Solaris is not known for that. Solaris is known for over engineering and reworking and reworking until the tech is good - which is why others look at Solaris to copy from.
"I started the Solaris Virtual Private Server Project in late 1999. By the end of 2000, we had a virtualized version of Solaris, rather similar to the Solaris 10 Zones feature."
And if we look at your link about different virtualization techniques:
We see that IBM WPAR is from 2007. And the "chroot" technique is from 1982 - it was created by a Sun co-founder.
This means I do not agree that IBM was first with Containers, and neither was Linux.
@Kraftman
Again, I only see Linux looking at other OSes and implementing their own version of a particular technique. Of course Linux is not copying everything verbatim, instead Linux devs rename things, and change here and there. But still, it is a copy of concepts and ideas. Linux is taking more than Linux gives back. Solaris and FreeBSD has invented much of Unix, and many OSes are using their techniques today. Including Linux.
I dont see cool Linux tech that other OSes are copying from. What would that be? Why would anyone copy BTRFS, which is a ZFS wannabe? Why would anyone copy KTrace or Systemtap which are inferior to DTrace? Everybody want DTrace and ZFS and Containers, including FreeBSD, Mac OS X, IBM AIX and Linux. I dont see everybody copying stuff from Linux. What has Solaris copied from Linux? Nothing?
Comment
-
Originally posted by kebabbert View PostEverybody agrees that IBM has done lot to virtualization, but that was back in the days, really long time ago when Mainframes were the hottest thing on earth. But not so more, most Mainframe innovation is done now. They are mature and not much happens in that field.<br>Originally posted by kebabbert View Post<br>
Regarding Containers, they are new to IBM, just as I said. IBM is calling it WPAR, and IBM implemented Containers in AIX v6.1 which came in year 2008. The IBM Container implementation is the latest of all OSes, and way many years after Solaris and Linux. So in a sense, IBM is copying others. Just as IBM Probevue is a copy of DTrace.<br>
<br>
From 2007:<br>
<a href="http://blog.thilelli.net/post/2007/05/22/Upcoming-IBM-AIX-6-features-vs-Sun-Solaris-10-and-OpenSolaris" target="_blank">http://blog.thilelli.net/post/2007/05/22/Upcoming-IBM-AIX-6-features-vs-Sun-Solaris-10-and-OpenSolaris</a><br>
"I can say I am impressed by both the upcoming POWER6 processor and the new AIX 6 release preview . They both are innovative, and come with a large set of new or updated features. For the purpose of this entry, I will concentrate on a little side-by-side comparison between the features announced for AIX, and those available in Solaris.<br>
The Workload Partitions (WPAR) on AIX seems to be very similar to the Containers, or Zones, which is an Virtualization feature found on Solaris."<br>
<br>
In year 2008, IBM says that WPARs is a <em>new</em> innovation, from the IBM web site (I did not know that IBM innovated Containers? Calling it "bold new innovation"?).<br>
<a href="http://www.ibm.com/developerworks/aix/library/au-wpar61aix/" target="_blank">http://www.ibm.com/developerworks/aix/library/au-wpar61aix/</a><br>
"WPARs are a bold new innovation, implemented within AIX 6."<br>
<br>
<br>
Originally posted by kebabbert View Post<br>
You are absolutely right that Solaris Containers is quite late, from 2004 (not 2005). <br>
<br>
But the Solaris Container concept is much older, but run under different names. Earlier, Solaris had something called BrandZ which is the same thing. And before that, Solaris had a similar technique where Solaris tried to run Linux apps in a container, called Lxrun. But Sun found out that implementation is not good enough so they reworked it once more, to improve it. For instance, Solaris Containers allow to install other OSes, which you can not do in Linux implementation. So when you look at the earlier attempts at Sun to do a good implementation of Containers, you need to go back to 1999, which is before Linux version. And as we all know, Linux looks at other OSes and implement their own version, but Solaris is not known for that. Solaris is known for over engineering and reworking and reworking until the tech is good - which is why others look at Solaris to copy from.<br>
<a href="http://www.kernelthread.com/publications/virtualization/" target="_blank">http://www.kernelthread.com/publications/virtualization/</a><br>
"I started the Solaris Virtual Private Server Project in late 1999. By the end of 2000, we had a virtualized version of Solaris, rather similar to the Solaris 10 Zones feature."<br>
<br>
<br>
And if we look at your link about different virtualization techniques:<br>
<br>
We see that IBM WPAR is from 2007. And the "chroot" technique is from 1982 - it was created by a Sun co-founder. <br>
<br>
This means I do not agree that IBM was first with Containers, and neither was Linux. <br>
<br>
Sun's static resources are basically LPARs which AIX had in 1986. <br> Of course this completely ignoring linux-vserver which had its beginning of life in 2001( <a href="http://www.solucorp.qc.ca/changes.hc?projet=vserver" data-cke-saved-href="http://www.solucorp.qc.ca/changes.hc?projet=vserver">http://www.solucorp.qc.ca/changes.hc?projet=vserver</a>). That seems well before Sun made containers a known project. So it seems disingenious to say linux has been simply copying Sun in this area.<br>
Comment
-
Jesus, this is unreadable. What did you do? Let me clean it up for you.
Originally posted by liam View PostI don't think IBM would agree with your statement but I'm not them, I don't particularly care about "Big Iron" and I'm not going to argue the point. You, however, do care about "Big Iron" since that is what Solaris' targets. Now, while it is true that this was done a long time ago that doesn't invalidate the point that the essential idea of virtualization goes back to, I would say, CP/CMS and they further developed the ideas over time. However you need to look at their System/3X0 machines not AIX. AIX was designed to compete with the cheaper UNIX stuff (actually, I think AIX is an official UNIX) and thus came to the party much later.
Not exactly. What you are referring to says they're new to AIX. AIX is not synomonous with IBM OS. Also HP-UX has had containers since HP-UX 9.0 (released in 1992).
Also according to this guy
Find out what distinguishes the top Unix versions in this comparison of IBM's AIX, Hewlett-Packard's HP-UX and Sun Solaris.
AIX's implementation of OS-level virtualization is better than Solaris'. So it was released a year later than Solaris (Solaris 10 being released in 2005, not the tech preview of 2004 which is why I gave 2005 as the year Sun intro'd it). Of course that doesn't say that IBM invented it first just that they were able to create a better version inside a year of Solaris 10.
That article contains lot of misconceptions and errors (or even IBM FUD some would say), see here which errors:
Do you want me to find an Oracle guy that writes "Containers are superior to IBM WPAR"? Hence, I dont trust your IBM source much. Do you have third party sources, such as researchers or someone else, that are more credible than IBM people? Likewise, I will not link to Oracle people because they are also a part in this, and are not objective. We must find someone from the outside with no affiliation to IBM nor Oracle, the best would be researchers. Or, I would accept an Oracle source that criticized Oracles solution - that is also credible. I would accept an IBM source that critiziced IBM solution - that is also credible. And I would accept an Linux developer that criticized Linux solution - that is also credible. But it is not credible when IBM critizice Oracle, nor the opposite.
Sun's static resources are basically LPARs which AIX had in 1986. Of course this completely ignoring linux-vserver which had its beginning of life in 2001
That seems well before Sun made containers a known project. So it seems disingenious to say linux has been simply copying Sun in this area.
Solaris Containers and IBM WPARs are the same thing (light weight). Solaris LDOM are similar to IBM LPAR (heavy weight). Why would IBM create WPARs and call it "new", if WPAR were similar to LPARs? No, LPAR and WPAR are not similar. Thus, Containers are not similar to LPARs.
Still I ask; when will we see Linux tech being copied to other OSes? Where is the new hot Linux tech that everybody wants and drool over? There is no Linux tech that I want. But Linux devs wants DTrace. Linux sysadmins wants DTrace. Linux sysadmins wants ZFS. They want Containers. etc.
Comment
-
Originally posted by kebabbert View PostJesus, this is unreadable. What did you do? Let me clean it up for you.
Regardless, apologies
Originally posted by kebabbert View PostYes, but Containers did not exist on Mainframes either. LPARs are far from it.
Originally posted by kebabbert View PostThis might be, I need to check this further later. But we can agree that Solaris is not copying Containers from Linux. Most likely it would Linux copying from others, as usual. As we know that Linux has looked at ZFS and implemented the best ideas from ZFS, and KTrace looked at DTrace, etc.
Originally posted by kebabbert View PostThis "guy" is an IBM supporter that writes most dubious things about Solaris.
That article contains lot of misconceptions and errors (or even IBM FUD some would say), see here which errors:
Do you want me to find an Oracle guy that writes "Containers are superior to IBM WPAR"? Hence, I dont trust your IBM source much. Do you have third party sources, such as researchers or someone else, that are more credible than IBM people? Likewise, I will not link to Oracle people because they are also a part in this, and are not objective. We must find someone from the outside with no affiliation to IBM nor Oracle, the best would be researchers. Or, I would accept an Oracle source that criticized Oracles solution - that is also credible. I would accept an IBM source that critiziced IBM solution - that is also credible. And I would accept an Linux developer that criticized Linux solution - that is also credible. But it is not credible when IBM critizice Oracle, nor the opposite.
You've already found oracle guys to quote. That's who you used when you were talking about AIX getting WPAR's and how they were similar to Containers.
The guy actually seemed pretty balanced I thought. He also seemed to know each system quite well. He gave props to each. Now, if you can point by point refute what he says with regards to PowerVM specifically "...No other flavor of Unix can boast these virtualization characteristics...".
Is it really so surprising IBM could do this? They've got like a quarter million employees, tons of money and they spend tons on research. I'm trying really hard to find out who invented the idea of containers but thus far I've yet to find a good source.
Originally posted by kebabbert View PostYes, IBM did a lot to virtualization earlier. But LPARs are old and LPARs is basically a reimplementation of LPARs from their old mainframes. That kind of virtualization has been around forever. Solaris copied that kind of virtualization, and the Solaris copy is called LDOM.
Solaris Containers and IBM WPARs are the same thing (light weight). Solaris LDOM are similar to IBM LPAR (heavy weight). Why would IBM create WPARs and call it "new", if WPAR were similar to LPARs? No, LPAR and WPAR are not similar. Thus, Containers are not similar to LPARs.
I'm also not sure WPAR's are exactly the same as Solaris Containers (though they do certainly overlap). However that's not the point.
I agree that LPAR's are different my point was that they are a neccessary step to OS level virtualization and that IBM had these long ago. I really don't care about AIX. Clearly it's not the high end IBM product. However, finding what exactly their z/OS can do isn't easy. IBM has tons of jargon and I'm not interested enough in mainframes to learn it:0
BTW, I noticed you didn't mention anything about HP-UX containers?
Originally posted by kebabbert View PostStill I ask; when will we see Linux tech being copied to other OSes? Where is the new hot Linux tech that everybody wants and drool over? There is no Linux tech that I want. But Linux devs wants DTrace. Linux sysadmins wants DTrace. Linux sysadmins wants ZFS. They want Containers. etc.
But I get annoyed when someone does those things.
Comment
-
Originally posted by liam View PostDo you have a source for that?
Ah, so you're using an oracle to refute this guy? Seriously? At least Ken was appearing to be objective (and frankly I'm not convinced that he wasn't). He's basically hyping Solaris and doesn't even pretend to be doing otherwise. Not a great source.
You've already found oracle guys to quote. That's who you used when you were talking about AIX getting WPAR's and how they were similar to Containers.
The guy actually seemed pretty balanced I thought. He also seemed to know each system quite well. He gave props to each. Now, if you can point by point refute what he says with regards to PowerVM specifically "...No other flavor of Unix can boast these virtualization characteristics...".
Just like when IBM copied DTrace and called it ProbeVue. IBM will never admit this, but even IBM supporters admit it is DTrace copy.
Is it really so surprising IBM could do this? They've got like a quarter million employees, tons of money and they spend tons on research. I'm trying really hard to find out who invented the idea of containers but thus far I've yet to find a good source.
Or the POWER6 cpus at 5GHz. IBM mocked Sun for Niagara cpus, which has many lower clocked cores. Back then, it was crazy to have 8 cores. IBM said that 1-2 cores in a cpu at very high speed, more than 5GHz, is the future. IBM said this, because "databases like strong cores, not many slower clocked cores" and mocked Niagara. And now we see the new IBM POWER7: it has many lower clocked cores. It certainly is not a single core cpu at 7-8GHz. Sun realized, that in the future you have to abandon the clock race, and switch to core race instead. Sun did that and IBM mocked Sun for that. And now IBM's POWER7 with 8 cores at 3.xx GHz is the best thing since sliced bread, according to IBM. Where are the 1 core CPUs at 8GHz? Nowhere. Everybody has now 8 cores. In a few years, there will be a 16.384 thread Solaris server which sounds crazy now, just as 8 cores sounded crazy back then. But I would not be surprised if everybody released such big servers later. Oracle shows the way, and the rest follows.
When Sun/Oracle do something, it is bad. Years later, IBM does the same thing and suddenly it is "new" and "innovative".
Have you seen the x4500 Thumper ZFS server? It is a 4U rack, and you open the top lid and lower the disks down in the chassi. The disks are standing up, and you just insert them down. It holds 48 disks, which is record for a 4U rack chassis. This is a new design, and people say it is innovative and new. Normally you just slid the disks into the front. Now Backblaze has been copying this design from Sun too. I would not be surprised if IBM suddenly releases a 4U rack chassis where you slide down the disks from the top of the roof, and call it "new and innovative".
So, yes, I would be surprised if IBM invented Containers. IBM did a lot of cool tech decades ago, but today? IBM has shifted from selling servers to be a more of a consulting company. IBM will even kill POWER servers sometimes soon. x86 servers are catching up fast on POWER servers.
POWER6 servers costed 5-10x more than x86 servers, and POWER6 servers were several times faster than x86.
POWER7 servers are only ~10% faster than x86 servers in some benchmarks, and POWER7 costs 3x more.
POWER8 will be slower than x86? And be even cheaper, otherwise people will not buy POWER. IBM does only high margin business, and walks away from low margin business. So, POWER will probably be killed. And POWER is the main platform for AIX. If POWER is killed, then AIX will be killed too. Incidentally, IBM has officially said that AIX will be killed:
ZDNET news and advice keep professionals prepared to embrace innovation and ready to build a better future.
I never said LPAR's weren't old, that was my point. IBM's had them in AIX since 86. Sun only got them about 10 years later.
I'm also not sure WPAR's are exactly the same as Solaris Containers (though they do certainly overlap). However that's not the point.
I agree that LPAR's are different my point was that they are a neccessary step to OS level virtualization and that IBM had these long ago.
I really don't care about AIX. Clearly it's not the high end IBM product. However, finding what exactly their z/OS can do isn't easy. IBM has tons of jargon and I'm not interested enough in mainframes to learn it:0
BTW, I noticed you didn't mention anything about HP-UX containers?
"This might be, I need to check this further later".
Perhaps. Solaris didn't release it's containers until well after linux-vserver. Do we have another source other than amit singh?
No idea. I don't particularly care. Linux is flexible (both horizontally AND vertically) and linux has had containers before Solaris (again, you failed to talk about virtuozzo and linux-vserver), but even if it didn't I don't care.
And Linux scales very good horizontally, yes. But vertically Linux scales very bad. There are no big Linux SMP servers on the market today. The largest Linux server I have seen is 48 cores - which is nothing compared to Oracles, IBMs and HPs offerings with up to 64 cpus and many more cores. Of course, in a cluster Linux scales very well, up to 10.000 of nodes. For instance, Google is said to have 900.000 servers
There are no SMP servers that big. It is a cluster. All computers bigger than 64 cpus are a cluster. The biggest Linux server - that is not a cluster - has 48 cores as I have seen. But I am not familiar with Linux servers.
If is a very useful system and moves at a very fast pass. The decision making is transparent and it has incredibly wide and varied uptake. However, even if it didn't I don't care. I'm not interested in seeing linux "win". I don't own stock in "linux", I don't contribute to the kernel, and I don't feel a need to tell others that "mine is better".
But I get annoyed when someone does those things.
Stallman worked on GNU operating system for decades, and suddenly a teenager from Finland with freckles comes in and creates a lousy kernel and gets all cred as created a new OS called "Linux". Stallman criticized Linus for naming "his" OS after him. And Linus claims that "no, I did not name it Linux, someone else did". Yeah sure. Linus has a big ego and he would never allow anyone to dictate over Linus creation. I would not. And Linus big ego is well known.
I would not be surprised if Linux people will soon claim that BTRFS is the original and ZFS is a copy. Or all DTrace derivatives, that DTrace is a copy. But hey, Linux is built on stealing someone else's code and claiming it as his own. So why would I be surprised? Linux is built on unhealthy ground, stealing someone else's work - decades of work by Stallman and the other people contributing which is disrespectful to them.
Comment
-
Originally posted by kebabbert View PostAnd Linux scales very good horizontally, yes. But vertically Linux scales very bad. There are no big Linux SMP servers on the market today. The largest Linux server I have seen is 48 cores - which is nothing compared to Oracles, IBMs and HPs offerings with up to 64 cpus and many more cores. Of course, in a cluster Linux scales very well, up to 10.000 of nodes. For instance, Google is said to have 900.000 servers
There are no SMP servers that big. It is a cluster. All computers bigger than 64 cpus are a cluster. The biggest Linux server - that is not a cluster - has 48 cores as I have seen. But I am not familiar with Linux servers.
That's why Linux supports 4096 cores on a single-image system (along with several TB of RAM).
Originally posted by kebabbert View PostAnd I get annoyed when someone else gets the cred. Dont you? Assume you done all the work, and then suddenly someone else comes in and claims he did everything and gets all credit. I promise you would get annoyed too, when you dont get the cred you deserve.
Stallman worked on GNU operating system for decades, and suddenly a teenager from Finland with freckles comes in and creates a lousy kernel and gets all cred as created a new OS called "Linux". Stallman criticized Linus for naming "his" OS after him. And Linus claims that "no, I did not name it Linux, someone else did". Yeah sure. Linus has a big ego and he would never allow anyone to dictate over Linus creation. I would not. And Linus big ego is well known.
Yes, Linus did admit to naming all his software after himself--"First linux, now git".
On another point:
Originally posted by kebabbert
Why dont Linux developers invent something really cool that others want to copy? I have never heard of any Linux tech that is worth copying. Have anyone heard of cool Linux tech that everone wants? No? Linux developers are just copy cats. Also, FreeBSD users tend to agree with this. I would like to see the Linux developers invent something really awesome that Solaris and AIX would be copying, then I would change my view of Linux developers as copy cats. Until now, they are actually copy cats, they are copying more than they give back to other OSes. Do they give back anything at all?
Who did KMS (real DRM+KMS, not just frame buffers) first?
Didn't Brian Paul start Mesa to use OpenGL on Linux (though SGI had released the standard)?
Ever read up on the origins of XRandr (Keith Packard's extension to reconfigure X on the fly, so he could autorotate an iPAQ display with an accelerometer)?
Could you remind me when Sun implemented Wake on Wireless Lan?
Why is Linux emulation the de facto standard for binary compatability (eg, lxrun)?
Does any other kernel run natively in userspace (Colinux, User-mode Linux, linux a386)?
Comment
-
Kebabbert - you seem to love the sound of your own voice. There's a saying "I don't need anger management, you just need to STFU". This applies very much in your case. If Solaris were a blow-up doll, you would be pimping yourself out day and night.
Coming back to important facts - of which you don't have ANY in your posts - BtrFS is superior to ZFS due to pervasive usage of B+ trees. They may have similar features and hence high-level users like yourself may think that ZFS is the original, since it came earlier. However, BtrFS (named for B+ TRee FS) has a newer and more original approach, which lets it deliver ZFS features in a simpler and more elegant manner. Makes BtrFS more flexible as well. e.g. Can you convert ext4 to ZFS on the fly ? No. But you can convert ext4 to BTRFS on the fly. Why ? Please do some research on why...
I registered just to tell you to either talk like an Engineer or STFU. You fill your empty posts with vast amounts of meandering prose to hide the complete lack of any content. Please don't waste any more electrons - there's a kid somewhere in a developing country who needs it more.
Comment
-
Originally posted by Ibidem View PostSo you didn't know about the SGI Altix servers?
That's why Linux supports 4096 cores on a single-image system (along with several TB of RAM).
There are no SMP servers that big. It is a cluster. All computers bigger than 64 cpus are a cluster.
These huge Unix servers costs many million of USD and has been sold for decades. They have typically 64 cpus, at the most. There are no bigger SMP servers on the market today.
A question to you: why dont these big companies just insert 128 cpus instead of 64 cpus? Or, even 256 cpus? Or, even 1024 cpus? Why dont they do it? These companies fight fiercely for the best TPC-C benchmark world record, and other world records. Why dont they just insert many hundreds of cpus and claim the record?
Answer: These servers are SMP servers. Basically, they are one huge fat server. Read more on "SMP".
The opposite is basically, clusters. The Linux SGI Altix server is a basically a cluster, a bunch of nodes sitting on a fast switch. Here is an article about the new ScaleMP company that has a solution called vSMP that connects together as many as 128 AMD Opterons, running Linux:
Sure, it might run a single Linux kernel spanning all of the nodes, but that does not make it into true SMP. It is still a cluster.
Let me quote the article:
Or you can get vSMP from ScaleMP and use a bunch of smaller and cheaper two-socket boxes (or maybe even single-socket boxes) to create a virtual fat memory system.
The vSMP hypervisor that glues systems together is not for every workload, but on workloads where there is a lot of message passing between server nodes ? financial modeling, supercomputing, data analytics, and similar parallel workloads
Someone explains:
"I tried running a nicely parallel shared memory workload (75% efficiency on 24 cores in a 4 socket opteron box) on a 64 core ScaleMP box with 8 2-socket boards linked by infiniband. Result: horrible. It might look like a shared memory, but access to off-board bits has huge latency."
So, you are wrong. There are no big "SMP" Linux servers on the market today. Of course, there are lots of clusters running Linux, and Linux is very good at running clusters. But for one fat huge server, the biggest Linux server I have seen benchmarks on, is 48 cores. There might be bigger. From the article above:
If you need more memory and more I/O and CPU oomph behind it, you are outta luck. You have to either parallelize your workloads or move to an eight-socket or if you are lucky and can find one, a sixteen-socket Xeon box.
Ted Tso, ext4 creator, just recently explained that until now, 32 cores was considered exotic and expensive hardware to Linux developers but now that is changing and that is the reason Ted is now working on to scale up to as many as 32 cores. But Solaris/AIX/HP-UX/etc Kernel devs have for decades had access to large servers with many cpus. Linux devs just recently has got access to 32 cores. Not 32 cpus, but 32 cores. After a decade(?), Linux might handle 32 cpus too.
Thanks to Eric Whitney’s benchmarking results, I have my money shot for my upcoming 2011 LCAtalk in Brisbane, which will be about how to improve scalability in the Linux kernel, using the case study of the work that I did to improve scalability via a series of scalability patches that were developed during 2.6.34, 2.6.35, and 2.6.36 (and went into the kernel during subsequent merge window). These benchmarks were done on a 48-core AMD system (8 sockets, 6 cores/socket) using a 24 SAS-disk hardware RAID array.
Originally posted by Ted Tso...Ext4 was always designed for the ?common case Linux workloads/hardware?, and for a long time, 48 cores/CPU?s and large RAID arrays were in the category of ?exotic, expensive hardware?, and indeed, for much of the ext2/3 development time, most of the ext2/3 developers didn?t even have access to such hardware. One of the main reasons why I am working on scalability to 32-64 nodes is because such 32 cores/socket will become available Real Soon Now...
Linus named it "Freax". But the FTP admin put it in a "linux" folder, and that stuck. Read up, and you'll see that.
Yes, Linus did admit to naming all his software after himself--"First linux, now git".
Let's see:
Who did KMS (real DRM+KMS, not just frame buffers) first?
Didn't Brian Paul start Mesa to use OpenGL on Linux (though SGI had released the standard)?
Ever read up on the origins of XRandr (Keith Packard's extension to reconfigure X on the fly, so he could autorotate an iPAQ display with an accelerometer)?
Could you remind me when Sun implemented Wake on Wireless Lan?
Why is Linux emulation the de facto standard for binary compatability (eg, lxrun)?
Does any other kernel run natively in userspace (Colinux, User-mode Linux, linux a386)?
Can you name something that everyone drools over, and wants? For instance, ZFS, DTrace, etc. Something that is really hyped? I have never heard Solaris or IBM AIX gurus being excited over something that Linux has. Can you name something? But everyone is drooling over ZFS and DTrace, and either porting it, copying it or stealing it.
Comment
Comment