This article explores the virtualization features available to administrators across several UNIX hardware platforms. Discover what they have to offer and how their features compare to PowerVM.
This article explores the virtualization features available to administrators across several UNIX hardware platforms. Discover what they have to offer and how their features compare to PowerVM.
This article is riddled with fundamental inaccuracies. Most of this data has been available for over an year. Very poorly researched.
Sun just announced a 4 chip UltraSPARC T2+ system that has 32 cores and 512GB Memory. Each core runs 8 threads. A domain can be assigned to each of these. So a 4-way UltraSPARC T2+ system can have 256 Logical domains.
http://www.sun.com/servers/coolthreads/t5440/
Sun also has T2+ chip based 2 way and have much larger memory configurations than 64GB.
http://www.sun.com/servers/coolthreads/t5240/
Low end and Midrange with the introduction of T5440.
CPUs can be dynamically added and removed ever since the first release.
http://www.sun.com/blueprints/0207/820-0832.pdf
“All CPUs exposed by the hypervisor are referred to as virtual CPUs. On
platforms supporting logical domains, such as a Sun Fire T1000 and T2000
system, each of the cores of the system has four executing threads, represented
as virtual CPUs by the hypervisor. Thus, an eight-core Sun Fire T2000 server
would have 32 virtual CPUs able to be partitioned between the various logical
domains on the system. With this release of Logical Domains 1.0 software, virtual
CPUs are able to be dynamically reconfigured; that is, removed or added to a
guest logical domain while the guest operating system is running, without
requiring a reboot. Note this requires a specific version of the Solaris Operating
System to be installed in the guest domain and might not work with other
operating environments.
”
You can have 32 domains on T1, 64 on T2, 128 and 256 on T2+ based systems depending on the configuration.
Sun’s Logical Domains provides the exact same capabilities. Some how the author missed that while doing research as well.
What could have been a good comparison ends up being a typical IBM marketing FUD piece.
Edited 2008-10-15 16:02 UTC
We are using zones here, we have only problems.
– unability to patch zones (we are using u3)
– performance problems (thanks to processes block in kernel mode).
– and now the servers reboot itself in u5
Solaris 10 is the most crappy release from Sun. You are taking Sun’s blueprints/site like a reliable source. But it’s only some ads from Sun, in fact the reality is by far different.
On the other side AIX and LPAR are working like a charm…
Edited 2008-10-15 16:25 UTC
Thats interesting, as where I work is a primarily IBM site, and pSeries/AIX is the buggiest P.O.S I’ve ever worked on.
LPAR’s are not totally separate, I’ve seen DLPAR operations on one LPAR kill another due to memory conflicts. We frequently see Power5 and 6 boxes drop when they suddenly offline 8gb of RAM, or decide to offline a cpu for laugh.
The IBM VIO server is memory and CPU intensive and prone to having strange issues which you just can’t trace through its complicated configuration (why the hell is every network card enX on AIX???).
I guess its what you’re used to… I work on AIX and yearn for Sun – looks like you’re the other way around!
AIX has LPAR support since 2000 or 2001. Sun is trying a new thing every year, it’s annoying.
To be able to use LDOMs in our one-year-old Sun T1000s, We had to upgrade the OS, the firmware, and install ten freakin patchs.
That’s not what I call a “mature technology”…
Edited 2008-10-15 19:19 UTC
U3 is fairly old now.
Can you elaborate on this please? Is there a bugid?
bugid?
Can you point out which Blueprints are incorrect so
we can get them fixed please. It’s also interesting
that you use this argument as a retort – did you
read the article linked by the OP? What would you call
that?
charm
~Ac^AEUR^Ac noun
1 the power or quality of delighting or fascinating others.
2 a small ornament worn on a necklace or bracelet.
3 an object, act, or saying believed to have magic power.
Seems somehow appropriate given the nature of your
post.
Yes and we paid to be early adopters of solaris zones.
Performance problem is mainly due to the fact that each zones share the same kernel. Thus when a software goes in kernel mode, the CPU time is not controlled anymore and the zone. So if there is a software bug during kernel mode, the zone will eat all the CPUs despite ressource control set.
Its no bug, it’s the way zone are working.
http://forums.sun.com/thread.jspa?forumID=844&threadID=5316991
you can find the bug ID (6696124) in the thread but Sun removed it for security reason ….
Can you give a concrete example of how to reproduce this problem so it triggers the said kernel bug?
Are you hitting these issues or just making hypothetical claims based on the design?
The same could be said for WPARs too. That’s why Sun offers different virtualization technologies as does IBM.
The thread says a patch was delivered. The OP didn’t post again so presumably it worked for them. Since they kept posting that they were unhappy, silence implies they are happy now.
I fail to see your point there. Are you saying AIX doesn’t have bugs? I am sure a little searching will provide me with enough bugs where AIX boxes crash and reboot too.
Edited 2008-10-16 16:09 UTC
If Sun hadn’t removed the bug report for reasons of probable paranoia, maybe we would actually know what had happened and what the fix was. I assume that’s what bug tracking systems are for.
If you’re suggesting that people should merely wait until they get silence to assume something is fixed then I’m afraid that’s rank amateurism at best, and that’s reason enough to question usage of a technology – especially if you’re paying for it.
Well … i was still able to look up the BugID in sunsolve.sun.com . And by the way: There is nothing as too much paranoia in security. Just because you~Ac^AEUR^ATMre paranoid, doesn~Ac^AEUR^ATMt mean they~Ac^AEUR^ATMre not after you.
Ignoring the sarcastic tone in the above post (as we do not use sarcasm here in the Middle East) the meaning is given here: http://idioms.thefreedictionary.com/work+like+a+charm:
work like a charm
if a plan or method works like a charm, it has exactly the effect that you want it to.
or
to operate perfectly.
Edited 2008-10-16 18:42 UTC
While I’m generally neutral on this topic, the summary at the bottom tells me all I need to know about the article. I might have expected better from IBM Developer Works:
Why this could not be true ?
Did you try solaris zones ? Sun LDOM ? IBM LPAR ?
LDOM is a new product, and they are just saying that IBM LPAR are here and *working* for years dude. Sun is late, it’s a fact.
cmon. this is an IBM commercial.
while the whole commercial was entertaining, the thing that made me laugh was this, “IBM has a 40-plus year history of virtualization. No other vendor can come close to making this claim…”
now at first there isnt anything really funny about it until you consider this. IBM has only updated its virtualization software AFTER another vendor did (Sun, HP, OSS, etc…). IBM released a product and then plays catch up to every comopany’s new release, the “me too” game. not that i don’t like there virtualization, infact i love WPARs, but its still nothing to write home about.
Are you kidding ? Which catch up ? Sun is releasing LDOM 5 years after IBM released LPAR taken from mainframe.
WPAR are useless on AIX as LPAR are so advanced.
Really? IBM has had virtualization technology since the late 60s. First on the mainframe platform starting with the OS level with CP/67 which became VM/370, VM/SP/ VM/ESA and today z/VM, and then as a Hypervisor in microcode as PR/SM and LPARs. This technology has trickled down to AS/400 (now System i) and now recently to System p.
Seems like a different person wrote the HP-UX part, which was pretty clear and understandable, and the the AIX and Solaris parts, which were inconsistent and confusing. It seems like the Solaris/AIX person didn’t understand the difference between Containers and LPARs, and, if I understand WPARs right (shared kernel, yes?), made the same mistake between WPARs and PowerVM hypervisor (obviously, if PowerVM can run OS400 and AIX, no shared kernel…or do you need a hardware partition for that? Wish the article would have told us.)
Plus, who copied and pasted from the powerpoint?
Anyway, great idea for an article, lame execution. This guy seems to know more about WPARs:
http://www.ibmsystemsmag.com/opensystems/december07/coverstory/1860…
Mr. Milbergs article is of abysmal quality. It’s partly based on outdated informations and otherwise outright incorrect. His position about partitions is based on the DSDs, not mentioning that the Quad-XSB introduced with SPARC Enterprise are a different story. The LDOMs stuff is based on T1000 systems to a part and outright incorrect at other parts. Mr. Milberg should read the documentation available at docs.sun.com before writing such an article to get the facts right.
I’ve already wrote a comment to this article in my own blog: http://www.c0t0d0s0.org/archives/4931-Analysing-a-so-called-Compari…
I don’t know how this article was able to find it’s way into IBM developerworks or even OSAlert.
Edited 2008-10-16 10:54 UTC
This data is also quite outdated and wrong regarding HP systems. For example:
– You can move resources between nPartitions using iCAP
– IBM has no electrical isolation, whereas on HP systems you can replace a whole cell board without downtime. And no, you don’t need to reboot for that.
– HP’s virtual machines product is called HPVM or HP Integrity Virtual Machines, IVM is an IBM product.
– HPVM supports 8-way guests today (not 4), runs on systems with up to 128 CPUs and I believe 2 TB of memory.
– HP demonstrated on-line migration, it’s currently in restricted beta with the latest release.
I’d like also to point out that HP virtual machine technology is software, so you can run it on older hardware. In other words, someone who purchased HP equipment 3 years ago will get on-line migration on their existing systems, but you need to buy new Power systems to get the equivalent feature.
Is it just me or are all of the IBM DeveloperWorks articles on OSAlert of middling to poor quality? The ones about linux are usually fairly superficial and the others seem to read like IBM brochures aimed at the less-than-sophisticated.