Response to "A comparison of virtualization features of HP-UX, Solaris and AIX"
By Jsavit-Oracle on Mar 25, 2010
Oh no, not again...
I've been able to avoid refuting FUD and misinformation for quite a while (and stick to my preferred topic: blogging on technology), but I just got an e-mail pointing me to Ken Milberg's A comparison of virtualization features of HP-UX, Solaris and AIX on IBM's website, and have to respond.
Just as in his previous article, to which I responded 3 years ago in Response to IBM, Sun and HP: Comparing UNIX Virtualization Offerings, he "surveys" the Unix vendors virtualization technologies and makes observations about different virtualization products. Many of his comments are simply wrong. As before, I'm not going to comment on his remarks about HP, for the simple reason that I don't know HP technology well enough, but I will correct mistaken remarks about Sun (now Oracle) capabilities. Note: his article is updated as of March 23, 2010, so it should be up to date in content.
Some miscellaneous bits first
Mr. Milberg says "Sun is also claiming features such as predictive self-healing, which has long been available on the System p." It's not a "claim" - it's a delivered feature of Solaris 10 for several years. We can decommission parts before they fail, when they show signs of getting sick, as well as recover from errors. I don't believe AIX has anything comparable. Perhaps he should take a closer look to understand what Solaris does - whether on SPARC, Intel, or AMD (something the "p-only" AIX cannot do.)
He also refers to VirtualBox with a reasonable description - but omits the fact that this is a capability that we at Oracle have that IBM lacks - a desktop virtualization product that runs on the most popular chipset in the world. To anybody who wants to run multiple virtual machines on their PC or laptop, don't hesitate: go to http://www.virtualbox.org/ and download a copy. It runs on Oracle Solaris, Linux and Windows, and can host guests running those operating systems, BSD, and even OS/2!
Virtualization, point by point
First, the names: even before being acquired by Oracle, Sun no longer called our virtualization products "xVM". Now, of course, they are part of the Oracle Virtualization product set and are named accordingly - Logical Domains are now "Oracle VM for SPARC." This is something an article updated 2 days ago should reflect. But that's a minor item when there are more serious factual errors to deal with.
He spends some effort disparaging Logical Domains, which I correct in this table. Note that he insists on calling things "partitions", when they're not: that's an IBM-specific term which doesn't apply elsewhere.
|"Scalability - Only eight CPUs and 64 GB RAM on one machine"||Wrong by a mile: a T5440 goes up to 256 CPU threads on 32 cores on 4 chips, and goes up to 512GB of RAM|
|"Server-line - Only low-end Sparc servers are supported"||Wrong: Machines like a T5440 or T5240 are nobody's "low-end" machine|
|"Limited micro-partitioning - Four partitions on T1, 8 on T2"||Boy, oh boy is this wrong. A T1-based server (no longer sold) could go up to 32. T2-based servers can go up to 64. On the T5x40 servers based on the T2 plus chip, you can have up to 128 domains.|
|"No dynamic allocation between partitions"||Wrong again: you can transfer CPUs, cryptographic accelerators, and I/O assets non-disruptively. Using the free resource manager in LDoms 1.3, you can even move CPU capacity between domains automatically, based on resource requirements.|
None of the statements he makes above are correct - he is wrong about fundamental platform capabilities. Contrary to LDoms having "many inherent flaws", they are a popular no extra cost feature of Oracle's Sun SPARC T-series product line, that compare very favorably to the expensive and less flexible virtualization options available on IBM POWER.
Besides these fundamental errors, Milberg misses some crucial points: He forgets to mention is the massive license fees you have to pay to use virtualization on POWER: Power7 licensing pricing for AIX, PowerVM and SWMA (required Software Maintenance) are extremely expensive: AIX 6.1 on a POWER 780 8/32 is about $130K, and PowerVM Enterprise Edition $89K. That is a pretty hefty price - exclusive of maintenance! Did I mention that Logical Domains, aka "Oracle VM for SPARC" comes with no cost at all?
Further, to do crypto at hardware speed on POWER, you have to buy crypto accelerator device$. Also, while Milberg makes a bunch of comments about mobility - he forgets to mention that domains can be moved between T-series servers without a reboot. That's another no-added-cost feature.
Finally, he makes a number of comments about Oracle Solaris Containers including the (fairly accurate) comment that we "had it and IBM did not." Well, that's only half the story: IBM and its proxies like Ken spent several years disparaging Containers - until they imitated them! :-)
Ken Milberg's article claims to be a comparison of virtualization technologies, but is marketing posing as analysis, and is full of fatal errors.
The article trots out the old chestnut "IBM has a 40-plus year history of virtualization. No other vendor can come close to making this claim.", which would be interesting if any of the POWER technologies were based on VM/370... but they're not. He pulls out the second chestnut he's used before "They offer one virtualization strategy, PowerVM, unlike the myriad of solutions available from Sun or HP", which is weird in an article that not only names several virtualization technologies available on POWER: PowerVM and WPARS, but also refers to the (completely unrelated) mainframe virtualization technologies. In reality, IBM offers multiple virtualization strategies - which isn't a bad thing (we do), but it's contrary to Ken's comment. Unfortunately for IBM, they don't have their own products on x86 servers, so their solutions depend on 3rd parties, while Oracle has Oracle VM, Oracle VirtualBox, and of course Solaris Containers - providing a complete virtualization portfolio.
In short, his article is merely a pitch, and is replete with errors. Readers who want an accurate comparison of virtualization technologies will have to go elsewhere.