Post to Twitter |
Comments (49) |
Views (86976) |
How big is a 32MB hypervisor?
Obviously, the size that's important is the memory footprint of the virtualization software when it's running.
Apparently, 32MB is 415MB in size, based on the attached screenshot from VMware ESXi 3.5. This is a 4GB Dell 2950 -- one of the two models on which ESXi Installable is, er, installable. While there is one virtual machine created, it's powered off.
And, as you see, it's consuming 415MB. Which is approximately 13 times 32MB.
Must be the New VMath™.
EDIT: It's been pointed out that in the last few weeks, VMware has been explicit about changing the ambiguously-worded "footprint" to "disk footprint." This moves the discussion from the realm of the interesting-but-inaccurate (a tiny memory footprint might indeed have some advantages) to the level of a stunt, and an uninteresting one at that (I have a t-shirt with more than 32MB of memory -- the cost difference between the 32MB USB keys used for trade-show giveaways and the 1GB USB keys used for real distributions of XenServer Embedded and ESXi is of negligible impact to any real server platform). So, oops, and, so what.
Comments (49)
Aug 05, 2008
Anonymous says:
Care to post the same for XenServer? Just to be fair...Care to post the same for XenServer? Just to be fair...
Aug 05, 2008
Roger Klorese says:
As soon as I have access to a freshly installed copy on a 4GB server, sure -...As soon as I have access to a freshly installed copy on a 4GB server, sure -- which will probably be tomorrow.
Although what are we being fair about? I'm not playing size comparisons -- only indicating that "32MB" doesn't mean what a reasonable user would expect it to mean in the ESXi hype, if it means anything at all. We've made no such claim about XenServer footprint.
Aug 06, 2008
Roger Klorese says:
See below. 324MB.See below. 324MB.
Aug 06, 2008
Andrew Storrs says:
Thanks Roger, that makes the comparison on base memory usage more interesting (I...Thanks Roger, that makes the comparison on base memory usage more interesting (I was the original commenter but forgot to login). I agree that the confusion (on the part of clients) related to memory usage by different hypervisors isn't helped by VMware's marketing group (how often do marketing groups help in that area anyway?
). But I would still prefer a flash bootable option for my hypervisor as it makes local storage or booting from a SAN less of a concern (which is great if you doing blades).
Aug 06, 2008
Roger Klorese says:
Understood. XenServer Dell Express and Dell Enterprise Editions are diskle...Understood. XenServer Dell Express and Dell Enterprise Editions are diskless-bootable now; HP Select and HP Enterprise Editions either are or will soon be.
Aug 06, 2008
Richard Snyders says:
This probably acutally matters for XenServer because it lacks any ability to ove...This probably acutally matters for XenServer because it lacks any ability to overcommit memory. But then Roger already knows about this and considers it also a VMware stunt.
Aug 06, 2008
Roger Klorese says:
As I posted, I don't consider memory overcommit to be a stunt -- I was o...As I posted, I don't consider memory overcommit to be a stunt -- I was one of the test users in Carl's original paper, with 38% savings on lightly-loaded Linux guests.
I consider wild claims of useful performance with reasonably-sized virtual machines and overcommit levels hyped as high as 2:1, 3:1 -- even 7:1 -- to be a stunt.
Aug 05, 2008
Anonymous says:
It would be helpful to link to VMware's claim that ESXi has a 32MB footprint in ...It would be helpful to link to VMware's claim that ESXi has a 32MB footprint in physical memory. What has been claimed from everything I have read is that it has a 32MB code footprint (of which it actually does). Applying the logic in the above post to Microsoft Windows XP would mean that since windows XP takes up the majority of a 750MB CD that it should use 750MB of physical memory which would not seem like sound logic.
Aug 06, 2008
Roger Klorese says:
It's funny, but... If your understanding is that the 32MB claim refers to the si...It's funny, but... If your understanding is that the 32MB claim refers to the size of the download or the region of the flash used by the static code, you may be right, but that is not the understanding of a single one of the customers or partners I've spoken to since ESXi was announced last VMworld (at which, by the way, we were showing XenServer Embedded to customers under NDA and at which time we'd already signed one or two OEM agreements).
And, indeed, you're right: VMware has taken to liberal use of the phrase "disk footprint" in the very recent past. But before the past couple of weeks, the only term used was "footprint" -- and with references to the "attack surface" in the security pitch, well, bits on disk don't have anything to do with "attack surface."
If that is the intent, it's not well-understood -- for example, the quote here that "My current (also free) VMware Server machine at work is running Windows Server 2003 x64. You can bet it has a much larger than 32 MB footprint! So, if I can rebuild that box as ESXi, I can probably host a few more virtual machines." A few more virtual machines, by saving a couple of gig of disk? Highly unlikely.
And here, both commenters assumed the 32MB figure referred to memory footprint as well.
Again: *every* *single* *person* *I've* *spoken* *to* in the past eleven months assumed that the number referred to memory consumption -- when we tried to describe the on-flash size of XenServer Embedded, they came back and said, "But that's not important -- how much memory does it take up? ESX 3i takes up 32MB."
If it's really just about media utilization and download time, well, pipes are fat, flash drives are cheap, and nobody cares. I'd stand corrected, but the information makes me want to lie down and greet it with a hearty "whoopie-doo."
If it costs a little bit of media space to take advantage of code that has withstood the security test of thousands of eyeballs, instead of black-box secrets... it's worth every byte.
Aug 07, 2008
Anonymous says:
Well I'm afraid I'll have to admit to falling for this marketing line. &nbs...Well I'm afraid I'll have to admit to falling for this marketing line. I'll also admit to not being an expert on virtualization, and as such not being in a good position to question my own interpretation of the message, be that as it may, I fell for it.
Aug 06, 2008
Anonymous says:
Umm, I don't work for vmware but I'm pretty sure the 32MB figure refers to the d...Umm, I don't work for vmware but I'm pretty sure the 32MB figure refers to the disk footprint of the hypervisor
Aug 06, 2008
Anonymous says:
Another well informed post (I also do not work for VMware). Exactly what point a...Another well informed post (I also do not work for VMware). Exactly what point are you trying to prove here Roger? VMware has _never_ said that the memory footprint was 32MB. In fact I'm surprised you havent picked up on the fact that the 'installable" edition actually takes up a few GB of disk space as it contains filesystem space for VMware tools (for the guests) and the VI client.
So if your main point is the memory overhead associated with the virtualisation platform, how about you post up some numbers regarding the memory footprint (and required diskspace) of Dom0? the Xen hypervisor is pretty useless without Dom0, how about we compare apples with apples?
Aug 06, 2008
Roger Klorese says:
As I already said: my main point was not comparison. (The memory footprint...As I already said: my main point was not comparison. (The memory footprints are similar.) VMware has never SAID that the memory footprint was 32MB -- they only implied it and allowed the perception to persist until a few weeks ago. Customers were messaged about "footprint" -- especially in regard to "attack surface" -- for nearly a year, with widespread understanding that this was memory footprint, not disk footprint -- an understanding for which I provided multiple neutral references. Since disk footprint means next to nothing,
Aug 06, 2008
Anonymous says:
The disk footprint means everything in this case because it means the Linux cons...The disk footprint means everything in this case because it means the Linux console was removed resulting in the smaller attack surface you continue to mention. That plus the ability to be embedded is entirely the point of noting the 32MB disk footprint.
Sep 04, 2008
Anonymous says:
In Roger's screenshot from XenCenter note that I believe the figure given for 'X...In Roger's screenshot from XenCenter note that I believe the figure given for 'Xen' memory usage does include dom0 usage, so the comparison was fair...
Aug 06, 2008
Anonymous says:
I have worked with VMware products for several years (I also do not work for VMw...I have worked with VMware products for several years (I also do not work for VMware), and all of the marketing that I have seen surrounding the ESXi product boasted the 32MB disk footprint, as it allowed server manufacturers to more easily embed the code into a chip on the system board, and thus no installation required. In the seminars/webinars that I have attended, this topic came up quite often, and the instructors made it pretty clear that the 32MB was not related to memory, but simply the hypervisor code.
Aug 06, 2008
Roger Klorese says:
While the 32MB number might make it easier for some maker of $25 appliances to e...While the 32MB number might make it easier for some maker of $25 appliances to embed ESXi, it has made no difference to makers of real computers. Those embedding it are using 1GB flash keys -- which makes the "accomplishment" pretty irrelevant.
Aug 06, 2008
Anonymous says:
The intent of my response was not to tout the cost difference between 32MB and 1...The intent of my response was not to tout the cost difference between 32MB and 1GB of flash, because I agree that the difference there is negligible. I rather pointed out that all of the media hype that I have seen/heard clearly focused on that topic, and as such made it evident that they were talking about disk footprint, and NOT memory.
Aug 06, 2008
Anonymous says:
I just wanted to make it clear that I also do not work for VMware. I have ...I just wanted to make it clear that I also do not work for VMware. I have nothing to add, just thought I should let you folks know that while I read this thread, I too am a non-VMware employee.
Aug 06, 2008
Roger Klorese says:
Congratulations. Say it proudly, as I can these days. (VMware former co-...Congratulations. Say it proudly, as I can these days.
(VMware former co-workers: kidding.)
Aug 06, 2008
Anonymous says:
It's pretty sad when the only thing you can find to nitpick about a competitor i...It's pretty sad when the only thing you can find to nitpick about a competitor is this. If "everyone you've talked to" thinks the 32MB is memory footprint, then perhaps you need to start talking to a few more technical people because no one that I know that works in the field considers the 32MB to be the memory footprint of the OS. I hope this doesn't reflect on the technical makeup of your coworkers at Citrix...
Aug 06, 2008
Roger Klorese says:
Did you bother to follow the links in my earlier reply? I posted links to ...Did you bother to follow the links in my earlier reply? I posted links to a couple of sources that clearly have this perception; I can link more if you want.
Aug 06, 2008
Anonymous says:
I'm not sure additional comments from people with the same lack of technical und...I'm not sure additional comments from people with the same lack of technical understanding constitutes a good defense of the point. Sorry, you can quote all the people you want that share the same misconception but that's all it is, a misconception on your part. Rather than arguing "that's what everyone thinks!", you might just want to fall back. Your article wasn't hammering the point that VMware's advertising on this point was misleading or confusing, but rather that the memory footprint wasn't 32MB. The footprint never was 32MB and while I appreciate you trying to save face and argue that it's all VMware's fault for misleading everyone, you should probably just fall back and leave this one alone. And VMware's position is quite clear on their website (http://www.vmware.com/products/esxi/) where it clearly says "disk footprint"
Aug 06, 2008
Roger Klorese says:
I could change this post now, but that doesn't change what it said yesterday. S...I could change this post now, but that doesn't change what it said yesterday.
Similarly, VMware's website does say "disk footprint" -- but it didn't a few weeks ago, or a few months, or when they announced it. It said "footprint," and made arguments like "attack surface" that are either inadvertently misleading or deliberately intended to imply memory footprint.
Search the web for "ESX 3i" and "footprint" and see what you get. (ESX 3i" will get you material older than a couple of weeks.)
Aug 06, 2008
Anonymous says:
Or it is possible the attack surface refers to the console-less-ness (is that a ...Or it is possible the attack surface refers to the console-less-ness (is that a word?) of the product rather than the non-existant 32MB memory footprint? Or it is possible that the footprint fell in line with their ability to distribute ESX3i on a USB thumb drive at VMworld? There are plenty of articles I can link to also that talk of the 32MB footprint in conjunction with the removal of the 2GB console refering completely to disk size. Mr. Petri doesn't seem to share the widespread confusion about the footprint back in November of 2007...
Aug 06, 2008
Roger Klorese says:
It would be possible for the "attack surface" language to refer to that -...It would be possible for the "attack surface" language to refer to that -- if they'd said it. But instead, what they said in their announcement was:
"The slim 32MB footprint of ESX Server 3i is a fraction of the size of a general-purpose operating system, resulting in a smaller attack surface"
...which is pretty clearly stating that "32MB == smaller attack surface."
Aug 06, 2008
Anonymous says:
Nothing in that statement refers to memory. The smaller attack surface is ...Nothing in that statement refers to memory. The smaller attack surface is solely a result of the Linux console being removed which does remove a wide range of attack vectors for the installable OS. You're not really doing yourself any service here by continuing this.
Aug 06, 2008
Roger Klorese says:
You could probably get down to 32MB without removing Linux, and remove Linux wit...You could probably get down to 32MB without removing Linux, and remove Linux without getting down to 32MB. The statement clearly says that the specific act of getting to 32MB reduced the attack surface, not side-effects of the steps along the way. The disk footprint has no direct correlation to the attack surface.
Aug 06, 2008
Anonymous says:
The fact that don't understand how removing 2GB of applications and a Linux...The fact that don't understand how removing 2GB of applications and a Linux console from an installable OS reduces the attack surface makes this entire thread even more pointless.
Aug 06, 2008
Roger Klorese says:
Taking well-inspected code with minimal services and replacing it with a secret ...Taking well-inspected code with minimal services and replacing it with a secret black box with proprietary code listening on most of the same ports is a security-by-obscurity argument.
XenServer Embedded is well under half a gig, and got there without throwing away the inspection value of open source implementations or losing the ability to run the hardware vendors' standard management agents.
Aug 06, 2008
Anonymous says:
Your ability to change your argument is astounding. Perhaps you should hav...Your ability to change your argument is astounding. Perhaps you should have written your original article on these points instead of having to constantly shift your defense in response to the comments here. I'm a customer of both Citrix and VMware and your interactions here are certainly not improving my opinion of your products or company.
Aug 06, 2008
Roger Klorese says:
Our core arguments -- that 32MB for its own sake is a pointless stunt, t...Our core arguments -- that 32MB for its own sake is a pointless stunt, that proprietary code results in more vulnerability than wide-open inspection -- have never changed. They were not the original point of this blog entry, but other people's comments give me ample space to reassert them.
The specific point of this blog entry -- that the 32MB that many understood to refer to memory footprint isn't correct when applied to memory -- is noted and acknowledged as being based on a misunderstanding, but one that VMware has fed via omission and misleading language up until a couple of weeks ago.
Aug 07, 2008
Anonymous says:
Now you are the one digging yourself into a hole. 32 MB is a direct refer...Now you are the one digging yourself into a hole.
32 MB is a direct reference to memory, MB being a universally understood shorthand fo computer 'memory'.
This the memory footprint is clearly stated to be 'fraction of the size of a general-purpose operating system'
and here's the payload
'resulting in a smaller attack surface'
So the small memory footprint is claimed to to result in a smaller attack surface.
For your argument to hold good - I would expect to see a statement as follows.
"The removal of the Linux console results in a smaller attack surface".
Got it?
Aug 08, 2008
Anonymous says:
It's all gone quiet...It's all gone quiet...
Aug 06, 2008
Anonymous says:
Umm, Seems like when you do something wrong and get caught, you argue louder and...Umm, Seems like when you do something wrong and get caught, you argue louder and stronger hoping that will get you out of the hole. The only thing it does is get you deeper in the sh*t hole. The above guy is right. I think you should just leave this one as it is. The more you try, the stupider will you look. No offense.
I am sorry if *everyone you know* or *everyone you talked to in the last 11 mo* thinks that you can run a server on 32 Mb. I suggest you find better company....
Aug 06, 2008
Roger Klorese says:
Better company than whom? Dozens of VMworld attendees? Hundreds of t...Better company than whom? Dozens of VMworld attendees? Hundreds of thousands of Citrix customers? Sorry, I don't think there is better company.
And I've had a couple of single-purpose Linux "servers" running on 32MB memory. It's perfectly possible. People who use the word "embedded" in its traditional sense -- like those who have demonstrated Xen running on a Samsung phone -- think 32MB is bloat.
Aug 06, 2008
Anonymous says:
You've only highlighted another misconception on your part. Embedded in th...You've only highlighted another misconception on your part. Embedded in the context of VMware would never refer to how small you can make things run in memory. A traditional embedded device would be something like a Netflix Roku player, but when talk of embedding VMware is made, it simply refers to the OS being small enough on disk to fit in a small flash RAM space making it easy to distribute with server hardware.
Aug 06, 2008
Roger Klorese says:
We announced embedded XenServer before ESX 3i was announced. We had agreem...We announced embedded XenServer before ESX 3i was announced. We had agreements to ship it before 3i did. We know very well what an embedded hypervisor means, in a server context.
We also know that the HP and Dell spec for hypervisor flash is a 1GB device, so there's no difference between a 32MB footprint and a footprint of 10x that in terms of server embedding. We take advantage of that space to store not only the hypervisor and control stack but also a rollback copy in case of flash update problems, as well as persistent configuration data.
32MB disk footprint? Interesting for trade-show giveaways, or if you're going to embed the hypervisor in a wristwatch or a Tamagochi. In an industry standard server? Not very interesting at all.
Aug 06, 2008
Anonymous says:
You're right. It isn't very interesting and isn't even one of the importan...You're right. It isn't very interesting and isn't even one of the important features of the product. Why you've decided to focus on it is beyond me.
Aug 06, 2008
Roger Klorese says:
First, this is one blog entry among dozens. I'd hardly call that "focus." ...First, this is one blog entry among dozens. I'd hardly call that "focus."
Second, if the original understanding of 32MB meaning memory footprint had been correct, and despite ten months of ambiguous VMware language it is now clear that it was not, the statistic would have been curious.
Aug 10, 2008
Anonymous says:
Roger, I completely disagree. It is rather easy to conclude that a 32MB disk foo...Roger, I completely disagree. It is rather easy to conclude that a 32MB disk footprint is in general going to have a smaller attack surface than something with 1GB or 1TB disk footprint. Generally speaking, more space = more code = more attack surface....that is using the term attack surface in the way that it is being used in this blog post. This doesn't imply than less disk footprint means better quality code, but you get the point.
Aug 06, 2008
Anonymous says:
Wow, you are saying you represent those poor "Hundreds of thousands of Citrix cu...Wow, you are saying you represent those poor "Hundreds of thousands of Citrix customers"?? All those down trodden customers who can't speak up; and you are their voice? Pleaseeee
I guess the best thing to do is NOT get the freebie ESX3i, but pay Microsoft for it's virtual product. I am sure that will solve all your problem.
Hey, who knows, maybe all this virtualizing is crap and no one should have got it.
BTW, "dozens of VMworld attendee think that VMware screwed them up with false propaganda? Umm, ok. And I was always taught that just because a lie is said 100 times doesn't make it true. The FACT remains, VMware has made it's stand clear on this issue. I guess by putting it on http://www.vmware.com/products/esxi/ they are saying, "Here it is. If you are still unhappy, sue me!!!"
Aug 06, 2008
Roger Klorese says:
Again: the word "disk" appeared on that page in the last three weeks. For ...Again: the word "disk" appeared on that page in the last three weeks. For the prior ten months, it appeared nowhere, and the supporting language about "attack surface" suggested the opposite. I guess you're right -- if we had only used the new Firefox clairvoyance extension back in September 2007 and read the July 2008 version of this page, there wouldn't be any confusion.
Aug 06, 2008
Anonymous says:
Since you like links, try this one http://blogs.technet.com/daven/archive/2007/...Since you like links, try this one
http://blogs.technet.com/daven/archive/2007/10/12/vmware-esx-server-3i.aspx#2268584
In the comemnts to this post from October of 2007, the commentor quotes the "original article" (which is the ESX 3i link) as stating clearly disk footprint.
And just as you noted, he did, that it really isn't an important issue, it still makes not of the fact that "the industry" isn't being misled here, just your lack of technical understanding.
Aug 06, 2008
Roger Klorese says:
True, he quotes the one web page where VMware used the word "disk," unlike the p...True, he quotes the one web page where VMware used the word "disk," unlike the press release and many other supporting documents that said only "footprint" until a couple of weeks ago. My profound apologies for not scouring the hundreds of references on their site and in the press to find this one gem and for not correcting everyone who had based their understanding on the other gazillion references.
Aug 06, 2008
Anonymous says:
Wow, I feel pretty smart now because I *didn't* think that the 32mb foot...Wow, I feel pretty smart now because I *didn't* think that the 32mb footprint was referring to memory. And this was based on my investigation of the product using the "misleading" literature prior to the site change. The "attack surface" references were clearly about fewer moving parts (software), and nothing to do with the memory.
Not everything is a conspiracy, maybe it's just a case of the original authors thinking something wasn't complicated and that people would understand. When some people were confused, they simply made a clairification for those people that didn't get it? Not as fun of an explaination, but probably closer to the truth.
Aug 07, 2008
Anonymous says:
A balanced, well thought-out and non-provocative response, by an anonymous poste...A balanced, well thought-out and non-provocative response, by an anonymous poster no less.
I highly commend you.
Yrs - the anonymous fool who really thought it was a 32 MB footprint.
Aug 19, 2008
Anonymous says:
I think this debate turned into (yet) another religious war..... I f...I think this debate turned into (yet) another religious war.....
I for first I have to admit 3i has been a great weapon in the hands of VMware marketing people (I have written about it months ago here: http://it20.info/blogs/main/archive/2007/09/19/48.aspx so I can't be referred to as a VMware bigot ...
).
Having this said this 32MB memory/disk footprint discussion is similar to discussing the quality of the peanuts at a wedding party where lobsters and fillets are being served in the other corner of the room.
My 2 cents.
Massimo.
Nov 08
Charm Lee says:
I think this 32MB memory discussion also is like a chain of pool supply struggli...I think this 32MB memory discussion also is like a chain of pool supply struggling against its competitor where different customers bid for lower prices. But 32 MB is not bad. I mean most have claimed so much about about the size of the disk footprint.. Im confused.
Add Comment