Bandwidth contention on the host's fiberchannel card (in most cases) is between traffic that's addressed to different physical disks in the SAN, meaning the different guests don't have to sit around waiting for each other's disk seeks to complete, the way they would if they had concurrent access to 2 virtual disk images on one physical disk.
Yes, but spindle contention has nothing to do with bandwidth (which is why even 1Gb iSCSI is rarely a bottleneck in most scenarios).
Also, it's quite common - particularly at the high end - to configure only a handful of LUNs/Arrays/RAID Groups/whatever, point multiple (dozens/hundreds) of VMs or servers at it and let the SAN controller intelligence and caching deal with the physical device contention.
The host's kernel can help a bit with disk caching and I/O scheduling/reordering to minimize the amount of disk thrash, but that's still a far cry from SAN performance.
If I attach a couple of dozen spindles to a local machine, it's going to deliver roughly the same (depending on the quality of the SAN and local RAID controller) performance as a couple of dozen spindles on a SAN.
We virtualized SQL Server 2008 R2 and ended up going back to Microsoft clustering. With clustering we still get HA but do not have to pay for VMware licenses. On VMware we were dedicating entire hosts to a single guest due to the high RAM utilization. In addition we were also taking the virtualization hit on the resource level by abstracting out disk and CPU access.
You're running clustered SQL servers, but VMware licensing was cost prohibitive ?
The "virtualisation hit" is minimal - a few percentage points. If it was a problem in a VM it would have been a problem on a physical box.
If the physical host has a lot of VM's using a lot of LUN's on the SAN, then there may still be contention for bandwidth on the fiberchannel card. Luckily this does not come with the massive overhead that is associated with contention for bandwidth on a local disk drive, but it's still a potential bottleneck to be wary of.
Saturating the bandwidth of even 4Gb FC, let alone 8Gb FC (or 6Gb SAS for the local equivalent) is quite uncommon.
Further, there's no reason bandwidth contention on FC has any less overhead than bandwidth contention on SAS, or even modern SATA.
The three biggest issues right now is Bandwidth, Disk I/O and memory;
I disagree on memory. Personally I find the increase in memory capacity to be the most amazing aspect of hardware growth over the last few years. You can put 1TB of RAM into servers these days, FFS. The smallest VMware hosts we buy are 256GB.
Similarly with bandwidth. 10Gbe and 8Gb FC provide more than adequate bandwidth for even the densest VM servers in all but corner cases. Heck, even 1GbE and 1Gb iSCSI is adequate for the fat part of the bell curve.
IO has, of course, always been a problem - but with more and more storage vendors finally starting to leverage SSDs for caching, it's becoming a solved (or at least mostly circumvented) problem.
For memory, you need MINIMUM 4 GB per VM, 6 if you want to play it safe. The hyper-visor alone costs 1GB or so to run. 8GB is the maximum memory size and most servers are tuned to 320GB per Processor max. If you wanted to virtualize 100 people, give them 60GB of disk apiece, plus 6GB of memory to run their apps in. You'd end up with a bunch of dual processor, 96GB of memory servers running about 13 users apiece, costing around 15k/year per server, plus licensing, plus the endpoint device cost.
WTF are you doing with VM specs like that ?
Typical office worker VDI VMs (Windows 7x 32) get 1.5Gb (maybe 2Gb) RAM each and thin-provisioning/linked clones should mean a per-user footprint of well under 10GB on your primary storage. Power Users get 3, maybe 4GB. Remember also that the typical environment can easily oversubscribe by 25% or more.
Broadly speaking, you can expect a consolidation ratio of about 4-8 VMs/core for a "normal" range of users and assume about 1Gb RAM each. This could go as high as 12-14 VMs/core (call centre) or as low as 1 (engineering/medical usage). So in the typical case a 48-192Gb RAM, 12-core Nehalem or better based server would comfortably run 48 - 96 VDI VMs at a server cost of US$5,000 - $10,000, depending on your storage connectivity.
If you're lucky enough to be doing a call centre style environment where requirements tend to be light, then 16 or 20-core, 256GB+ hosts should be able to comfortably exceed 200 desktops each.
The "problem" with VDI, like other terminal-style environments, is that you need large scale (1000s of users) to see real capex savings, particularly because the requisite storage infrastructure tends to be expensive. There are definitely manageability and opex savings at lower densities, but the up-front $$$$s need to be more closely examined.
That's funny, because in a recent Slashdot discussion about laptops the exact opposite was recommended - business grade laptops are typically priced higher for essentially the same hardware you get in the "consumer" grade.
It's not the hardware specs you're paying for, it's the better warranty and support and _vastly_ better case construction.
For at least three years the trackpads have been multi-touch; so, at least within bootcamp, the right click etc is emulated by having two fingers on the trackpad while clicking...
I no longer have Windows installed on my MBP to check, but it certainly used to be that the two-finger-right-mouse-button-emulation didn't work when trying to do a right-click-drag, which a surprising number of people use quite a lot.
Sounds like a job for the ACCC to me, they have the independence and the teeth to tackle something like this, as you say the opposition are not going to want to help the government look useful.
After ten years of Liberals in Government, the ACCC is but a shadow of its former self.
5 subsequent years of "New Labor" has (unsurprisingly) done little to remedy the situation.
Allow free importation of goods from the US and other markets and watch the vendor premiums for your mysterious island continent collapse. If Australians could simply buy from Adobe US, It'd be pretty difficult for Adobe to maintain a price premium...
There are few import restrictions to Australia in general and even fewer from the US thanks to the "Free" Trade Agreement.
i will sell everything i own and move to australia because it is the last western nation with a little redemption left in it
I wouldn't rush into it. Most of the people and politicians here seem determined to turn Australia into a clone of America. If the Liberals (Australian equivalent of the Republicans) take power in the next Federal election I, for one, will be aiming to leave the country.
The automobile can be a bit fuzzier - but certainly highway driving requires extreme amounts of attention. City driving isn't usually done for long stretches - unless it's stop and go, in which case nothing is happening to make it require much brain exercise.
This sounds backwards. City driving has a lot more hazards and variations that need to be tracked simultaneously. Trundling along on a highway at a pretty constant speed (probably using cruise control, at that) isn't especially taxing.
I would be interested to see some actual studies, but I would be surprised if the typical driver found highway driving more stressful than city driving.
First, show me where I said anything about a typical corporate environment.
That would be the part where you're making a sweeping, generalised judgement call. It seems reasonable to assume one of the most common scenarios would be encompassed.
Yeah. I'm funny like that. I think an order of magnitude increase in bandwidth has the capability to be more useful, and conversely an order of magnitude less could be considered "less useful." If 100Mb wasn't "less useful" there wouldn't be a 1 Gigabit standard.
However, this difference does not happen in isolation. You are trading off bandwidth against POE.
Ultimately, the question becomes: is 1Gb more useful or is POE more useful ? My answer is that in most common corporate environments, POE will be considered more useful because 1Gb is largely unnecessary.
I'll try to put it in a way you can understand. With Gigabit it has potential to do real computing... e.g. boot Linux with PXE and access a data store in a NAS, for example.
100Mb is quite adequate for this.
I don't care what it was designed for, nor did I claim that their design decisions were unsound. The whole point, which you are working so hard to not get (perhaps because you are too busy putting words in my mouth) is that anyone who had an idea of using this in some very cool applications with visions of high performance networking at their disposal and powering it with PoE is SOL.
Your original comment in this thread was: "... which drops to 10/100 when using PoE, thereby making it only marginally useful for very thin applications."
Which is patently false. 100Mb is not only very useful for just about anything anyone would want to do with a thin client, it's also quite adequate even for normal, managed desktop PCs booting from local disk and accessing data off the network. It's even adequate for thick clients booting over the network, as evidenced by all the places that not only did it before they could get gigabit, but continued to do it for years afterwards.
Fundamentally, the marginal utility of 1Gb over 100Mb for most end-user computing scenarios is very small. I know of several companies that have, within the last five years, replaced their entire office networks (multiple floors in multiple buildings, thousands of endpoints) and chosen to stay at 100Mb for ~95% of endpoints because no benefit (to justify the additional cost) was perceived in going to 1Gb.
Obviously HP put Gigabit capability in there for a reason (you did know that companies count every penny and add up the cost of the BOM, right?)
1Gb adds SFA to the cost of the end user device. It adds _shitloads_ to the cost of the networking infrastructure to support thousands of those devices.
It's kinda hard to say, given that it's not clear what you mean by "desirable". Have you considered Australia or New Zealand?
Both Australia and New Zealand are also still in the midst of massive real estate bubbles (Australia's, at least, is starting to deflate - though it still remains to be seen if we'll get a US-style ~4 year crash or a Japan-style ~20 year slow melt).
In fact, about the only places in the English-speaking world that don't have ludicrous real estate prices are Ireland and the USA.
You appear to be arguing that enabling POE renders the networking connection "marginally useful", or "less useful" because of a speed drop from 1Gb to 100Mb.
In the typical corporate environment, 100Mb is more than adequate even for a regular managed desktop PC, let alone a thin client. For nearly all thin client use cases, even 10Mb is quite adequate.
Or, to put it another way, your assertion is wrong. In a typical corporate environment, a 100Mb network with POE is almost certainly going to be *more useful* than a 1Gb network without POE.
No. You definitely don't get it. Hint: You take an order of magnitude hit in network performance if you power this with PoE. Something that was useful for much fatter thin clients now becomes only useful as a very thin client when you power it via PoE. That is the point in the context of the wonderful PoE capabilities of this box. Don't use PoE unless you don't want to take advantage of its full capabilities, because when you use PoE it becomes much less useful. If you don't like the word marginally, pretend it isn't there and think: much less. The point remains.
In most corporate environments POE is going to be far more useful functionality than another 900Mb of network bandwidth, since even a 10Mb connection would probably be quite adequate.
Coming from a Canadian that eventually moved to the UK.. At least over here the salaries and rates are commensurate with the property prices.. try living in Vancouver nowadays..
Hate to break it to you, but the UK is also in the grip of a real estate bubble. It slipped a bit during the GFC, but housing there hasn't really been affordable (median multiplier under 3) since the late '90s.
Vancouver, like Australia, is insanely overpriced. But that doesn't make the UK affordable.
No, I am inferring how the UK authorities think based on my understanding of their left-wing authoritarian mindset.
Only an American would call the current UK Government "left-wing".
The last time the UK had an actual left-wing Government was sometime back in the '70s. Like most of the Anglosphere, it's been moving further and further to the right for decades.
It'll be just like in the mid 1990's when Compaq switched the CD drives in their servers from SCSI models to IDE models because Microsoft told them to.
Rrrright. I'm sure the $hundreds it took of the price (thus making it easier for more people to buy) had nothing to do with it.
I thought I'd seen all the Slashdot conspiracy theories about Microsoft. Congratulations on coming up with a new one.
And it'll be just like in the late 2000's when Microsoft started forcing netbook manufacturers to lard up the specs on the previously cheap devices because they needed just enough horsepower to run Windows XP.
This is pretty cookie cutter stuff though. I'm sure the customer demand for machines running Windows had nothing to do with vendors trying to sell netbooks running Windows.
I'D RATHER THEY HAVE MORE MONEY TO HIRE PEOPLE. In case you haven't heard, jobs are scarce these days and I damn sure don't see the government hiring too many people.
Rest assured Google won't use that money to hire people.
You seem to disagree though, what restrictions (aside from criminal discrimination) would you place on a business removing employees?
A legitimate reason.
"Your job no longer exists" is a legitimate reason. So is "you're incompetent".
"We're going to fire you today so we can hire you back tomorrow to do the same work on half the pay" or "we're going to fire you so we can hire the new CEO's son into the same position" are not.
Slashdot Mirror
Yes, but spindle contention has nothing to do with bandwidth (which is why even 1Gb iSCSI is rarely a bottleneck in most scenarios).
Also, it's quite common - particularly at the high end - to configure only a handful of LUNs/Arrays/RAID Groups/whatever, point multiple (dozens/hundreds) of VMs or servers at it and let the SAN controller intelligence and caching deal with the physical device contention.
If I attach a couple of dozen spindles to a local machine, it's going to deliver roughly the same (depending on the quality of the SAN and local RAID controller) performance as a couple of dozen spindles on a SAN.
You're running clustered SQL servers, but VMware licensing was cost prohibitive ?
The "virtualisation hit" is minimal - a few percentage points. If it was a problem in a VM it would have been a problem on a physical box.
Saturating the bandwidth of even 4Gb FC, let alone 8Gb FC (or 6Gb SAS for the local equivalent) is quite uncommon.
Further, there's no reason bandwidth contention on FC has any less overhead than bandwidth contention on SAS, or even modern SATA.
You are thinking of FT, not HA.
(Which is easy to do, since really their names are back to front with regards to functionality.)
It was being done, but then Reagan and Thatcher happened, and it's been deteriorating ever since.
I disagree on memory. Personally I find the increase in memory capacity to be the most amazing aspect of hardware growth over the last few years. You can put 1TB of RAM into servers these days, FFS. The smallest VMware hosts we buy are 256GB.
Similarly with bandwidth. 10Gbe and 8Gb FC provide more than adequate bandwidth for even the densest VM servers in all but corner cases. Heck, even 1GbE and 1Gb iSCSI is adequate for the fat part of the bell curve.
IO has, of course, always been a problem - but with more and more storage vendors finally starting to leverage SSDs for caching, it's becoming a solved (or at least mostly circumvented) problem.
WTF are you doing with VM specs like that ?
Typical office worker VDI VMs (Windows 7x 32) get 1.5Gb (maybe 2Gb) RAM each and thin-provisioning/linked clones should mean a per-user footprint of well under 10GB on your primary storage. Power Users get 3, maybe 4GB. Remember also that the typical environment can easily oversubscribe by 25% or more.
Broadly speaking, you can expect a consolidation ratio of about 4-8 VMs/core for a "normal" range of users and assume about 1Gb RAM each. This could go as high as 12-14 VMs/core (call centre) or as low as 1 (engineering/medical usage). So in the typical case a 48-192Gb RAM, 12-core Nehalem or better based server would comfortably run 48 - 96 VDI VMs at a server cost of US$5,000 - $10,000, depending on your storage connectivity.
If you're lucky enough to be doing a call centre style environment where requirements tend to be light, then 16 or 20-core, 256GB+ hosts should be able to comfortably exceed 200 desktops each.
The "problem" with VDI, like other terminal-style environments, is that you need large scale (1000s of users) to see real capex savings, particularly because the requisite storage infrastructure tends to be expensive. There are definitely manageability and opex savings at lower densities, but the up-front $$$$s need to be more closely examined.
Reaching for what, exactly ?
I made mention precisely because I've been very surprised at people who do use it (ie: secretaries, CEOs, etc).
It's not the hardware specs you're paying for, it's the better warranty and support and _vastly_ better case construction.
I no longer have Windows installed on my MBP to check, but it certainly used to be that the two-finger-right-mouse-button-emulation didn't work when trying to do a right-click-drag, which a surprising number of people use quite a lot.
After ten years of Liberals in Government, the ACCC is but a shadow of its former self.
5 subsequent years of "New Labor" has (unsurprisingly) done little to remedy the situation.
There are few import restrictions to Australia in general and even fewer from the US thanks to the "Free" Trade Agreement.
I wouldn't rush into it. Most of the people and politicians here seem determined to turn Australia into a clone of America. If the Liberals (Australian equivalent of the Republicans) take power in the next Federal election I, for one, will be aiming to leave the country.
Eating less is quite difficult when you end up spending 3/4 of your day trying to ignore gnawing hunger.
This sounds backwards. City driving has a lot more hazards and variations that need to be tracked simultaneously. Trundling along on a highway at a pretty constant speed (probably using cruise control, at that) isn't especially taxing.
I would be interested to see some actual studies, but I would be surprised if the typical driver found highway driving more stressful than city driving.
That would be the part where you're making a sweeping, generalised judgement call. It seems reasonable to assume one of the most common scenarios would be encompassed.
However, this difference does not happen in isolation. You are trading off bandwidth against POE.
Ultimately, the question becomes: is 1Gb more useful or is POE more useful ? My answer is that in most common corporate environments, POE will be considered more useful because 1Gb is largely unnecessary.
100Mb is quite adequate for this.
Your original comment in this thread was: "... which drops to 10/100 when using PoE, thereby making it only marginally useful for very thin applications."
Which is patently false. 100Mb is not only very useful for just about anything anyone would want to do with a thin client, it's also quite adequate even for normal, managed desktop PCs booting from local disk and accessing data off the network. It's even adequate for thick clients booting over the network, as evidenced by all the places that not only did it before they could get gigabit, but continued to do it for years afterwards.
Fundamentally, the marginal utility of 1Gb over 100Mb for most end-user computing scenarios is very small. I know of several companies that have, within the last five years, replaced their entire office networks (multiple floors in multiple buildings, thousands of endpoints) and chosen to stay at 100Mb for ~95% of endpoints because no benefit (to justify the additional cost) was perceived in going to 1Gb.
1Gb adds SFA to the cost of the end user device. It adds _shitloads_ to the cost of the networking infrastructure to support thousands of those devices.
Both Australia and New Zealand are also still in the midst of massive real estate bubbles (Australia's, at least, is starting to deflate - though it still remains to be seen if we'll get a US-style ~4 year crash or a Japan-style ~20 year slow melt).
In fact, about the only places in the English-speaking world that don't have ludicrous real estate prices are Ireland and the USA.
You appear to be arguing that enabling POE renders the networking connection "marginally useful", or "less useful" because of a speed drop from 1Gb to 100Mb.
In the typical corporate environment, 100Mb is more than adequate even for a regular managed desktop PC, let alone a thin client. For nearly all thin client use cases, even 10Mb is quite adequate.
Or, to put it another way, your assertion is wrong. In a typical corporate environment, a 100Mb network with POE is almost certainly going to be *more useful* than a 1Gb network without POE.
In most corporate environments POE is going to be far more useful functionality than another 900Mb of network bandwidth, since even a 10Mb connection would probably be quite adequate.
Hate to break it to you, but the UK is also in the grip of a real estate bubble. It slipped a bit during the GFC, but housing there hasn't really been affordable (median multiplier under 3) since the late '90s.
Vancouver, like Australia, is insanely overpriced. But that doesn't make the UK affordable.
Only an American would call the current UK Government "left-wing".
The last time the UK had an actual left-wing Government was sometime back in the '70s. Like most of the Anglosphere, it's been moving further and further to the right for decades.
Rrrright. I'm sure the $hundreds it took of the price (thus making it easier for more people to buy) had nothing to do with it.
I thought I'd seen all the Slashdot conspiracy theories about Microsoft. Congratulations on coming up with a new one.
This is pretty cookie cutter stuff though. I'm sure the customer demand for machines running Windows had nothing to do with vendors trying to sell netbooks running Windows.
Rest assured Google won't use that money to hire people.
This is what's called a straw man argument.
The USA has amongst the lowest tax rates in the OECD. Particularly on the rich.
Reality, as usual, is in stark disagreement with loony right-wing rhetoric.
I don't.
It seems, therefore, we have a stalemate.
A legitimate reason.
"Your job no longer exists" is a legitimate reason. So is "you're incompetent".
"We're going to fire you today so we can hire you back tomorrow to do the same work on half the pay" or "we're going to fire you so we can hire the new CEO's son into the same position" are not.