From 32 bit to 64 bit

You’ve heard the drill, any system with 4GiB+ of RAM requires a 64 bit operating system. Why? Because the total addressing space of the memory (the number of locations on the physical memory of your computer) in 32 bit memory architectures is limited to a total of 4 GiB.

This can be calculated with the following formula: 2³², which equals 4,294,967,296 bytes, or 4 GiB.

Conversely, 64 bit has the following formula: 2⁶⁴, which equals a rather stunning 18,446,744,073,709,551,616 bytes, which translates into 16 EiB or 17,179,869,184 GiB.

So, it’ll take a while before we run out of addressing space in 64 bit memory architectures.

Anyway, you’ve probably recently heard about that thing called PAE, or Physical Address Extension. Most likely, you’ve heard it as a trick to make Windows XP recognize that extra ram, or your 32 bit Vista perhaps.

Is it true? Does it really work? The short answer is yes. It does work. However, we need to dive in the specifics a little bit to know why in the world Microsoft and other hardware/software companies decided to hide this Intel invention.

PAE Explained

When inventing the x86 architecture, and incidentally the x86-64 architecture (also known by its maker Advanced Micro Devices as AMD64), Intel also created something called Physical Address Extension (PAE) as a feature of the x86 architecture to make up for future memory limitations of the 32 bit architecture.

The trick is not a trick, it is a real hardware feature that consists of augmenting the number of memory address lines on the CPU from 32 bit to 36 bit, bumping the total possible amount of memory from 4 GiB to 64 GiB.

However, the processor remains a 32 bit processor, as well as its supporting motherboard and memory chips, preventing simultaneous use of more than 4 GiB. In other words, it remains possible to exploit the total 64 GiB of memory but no single virtual ram instance (ie. Photoshop) or physical memory unit can use more than 4 GiB of memory at the same time.

This translates into the impossibility of mounting anything else than 4 GiB sticks of RAM in your system, and probably even a total of 4 GiB because of the 32 bit architecture of your motherboard, which most likely only supports a total of 4 GiB of RAM, regardless of the CPU.

Now, since the total simultaneous limit is 4 GiB, you may be wondering how the operating system is capable of expanding this to 64 GiB. Effectively, since the x86 architecture is old and very common, it’s been long since operating systems including Windows and Unix variants like Mac OS X and Linux support the PAE extension, and with a few registry hacks, it’s possible to enable this “hidden” feature of Windows and other operating systems (in which case it’s something else than registry).

In Windows, this is called Address Windowing Extensions (AWE), where it involves mapping/spanning, or “Windowing”, an operating system across to more than a single virtual instance of 4 GiB of memory (or an application).

Why nobody told you

If you’re just a tad geeky in computers, you probably understood right away why software and hardware makers are pushing you towards 64 bit right away instead of temporarily using PAE.

And there, I’ve just said it, it’s a temporary solution. In a few years, systems will already be exceeding 64 GiB of memory, and applications (virtual memory instances) will need much more than blocks of 4 GiB of memory, a limitation of PAE.

Since 64 bit is a much more future-ready effort, instead of having to do two switches in a short period of time, Intel and Microsoft, among other companies, decided to literally hide this capability and make 64 bit the only apparent way to get more than 4 GiB of addressing space.

Should you use PAE

For the moment, if you’ve been using anything else than Windows, for instance, a Mac, the 64 bit issue should not have been an issue for you because prior earlier Apple processors like the PowerPC G5 (before the Intel x86-64 switch) were already 64 bit architectures.

Unlike Microsoft Windows and Linux, Apple’s tight integration of hardware and software allowed a much harder transition, from the IBM PowerPC to the Intel x86-64 microprocessor architecture, to be done in a very smooth way.

However, Microsoft’s much larger user base doesn’t allow this freedom and Windows XP remains a largely 32 bit operating system, seeing very limited support of its 64 bit counterpart, released in 2003, the same year Apple released its own first 64 bit system, the Power Mac G5.

To smooth in the transition, Microsoft also made Windows Vista and its more recent Windows 7 into two editions, 32 bit and 64 bit, although in this case, support is much more widespread, especially caused by the wider availability of 64 bit processors from Intel.

Along with Windows Vista came the mainstreaming of 4 GiB of system RAM and its related problems.

If you were hesitating to swith over Windows Vista 64 bit and found PAE to be a good solution to solve your RAM problem while keeping Windows XP, I would reconsider.

With the imminent launch of Windows 7 and Windows XP’s extended support period ending in 2014, PAE is but a very temporary solution.

How to

But hey, it can be useful where 64 bit is not possible because of a 32 bit processor, so why not. Here goes, this option is compatible with any Intel Pentium Pro, Pentium II, III, 4, Core, Core 2, Core i7 and + processor, along with every recent AMD processors and Athlon series.

Windows XP

1. Open an explorer window
2. Tools > Folder Options > View Tab
3. Check the radio box written “Show hidden files and folders”
4. Click OK to accept changes and close the dialog box
5. Go to your local drive where Windows is installed, most likely C:
6. Locate the file called BOOT.INI
7. Right-click on the file and click Properties
8. In the Properties dialog box, make sure the Read-only attribute is unchecked (checking it will prevent you from modifying the file)
9. Click OK to accept changes and close the dialog box
10. Open the BOOT.INI (default opens with Notepad)
11. It should look something like this:

[boot loader] timeout=30 default=multi(0)disk(0)rdisk(0)partition(1)\WINNT [operating systems] multi(0)disk(0)rdisk(0)partition(1)\WINNT=”Microsoft Windows XP Professional” /noexecute=optin /fastdetect

12. Append at the end of last line the following: \PAE
13. It should now look like this:

[boot loader] timeout=30 default=multi(0)disk(0)rdisk(0)partition(1)\WINNT [operating systems] multi(0)disk(0)rdisk(0)partition(1)\WINNT=”Microsoft Windows XP Professional” /noexecute=optin /fastdetect /PAE

14. If it does, save the file, exit Notepad and restart Windows

Congratulations, your Windows XP system now runs with PAE enabled.

Windows Vista / Windows 7

1. Click on the Start Orb
2. Search for CMD
3. Right-click CMD or Command Promt in the search results and click Run as administrator
4. In the command line, enter:
bcdedit /set pae ForceEnable
5. Close the command line

Congratulations, your Windows Vista/7 32 bit system now runs with PAE enabled.

To install PAE on a specific boot if you’re using a dual/multi-boot system, refer to the MSDN BCDEdit /set command documentation for instructions on how to set the ID of the boot.

Note: Official Microsoft documentation does not as of this writing specify explicit support for PAE in Windows 7.

Note: Applications not specifically built to use the AWE api from Microsoft will not be able to use more than 4 GiB of RAM even though PAE is enabled, this includes about every non server-specific application possible, so no, Photoshop and company won’t support more than 4 GiB of RAM each.

Note: Even though PAE technically enables 64 GiB of RAM, Windows XP, along with Windows Vista 32 bit, is limited to a total of 4 GiB. Detailed information can be found in this MSDN article on memory limits of Windows. Windows Vista 32 bit and Windows 7 32 bit also support PAE, as mentioned in other Microsoft articles spread amongst MSDN and other resources. However, Microsoft documentation isn’t clear on whether it is the physical RAM limit or the physical total addressable memory limit that is 4 GiB so results of enabling PAE may vary. With luck, they meant the total physical RAM limit, and not the total physical addressable memory limit, since this one is already 4 GiB with the 32 bit architecture, which would make PAE useless.

Note: Although the x86-64 (64 bit) architecture does support PAE, Windows’s AWE does not support it.

Note: Even though Intel’s specifications limit PAE to an extended total of 36 bit of addressable memory, it is technically possible to example this amount without changing the hardware. For example, Windows Server 2003 SP1 Datacenter Edition uses a special 37 bit PAE capable of supporting 128 GiB of RAM.

Mac and Linux

PowerMac processors are 64 bit since the G5, and any Intel Mac is 64 bit too, so you don’t have to worry about PAE at all, just the OS’s own RAM limitations (up to 16 TiB of RAM in OS X Snow Leopard 10.6). In Linux, any distro on the 2.6 version and up of the kernel has PAE included native and most often enabled by default, so it’s another thing you don’t have to worry about.

This often overlooked issue is easily solveable. Since the whole computer system is based on the SI system, which only standardized unit/time, every other option is a non-standard invention by some computer engineers.

So if you see people using anything else than kb/s, kB/s, they are wrong.

However, the IEC’s standards are not SI’s, so Kibit/s instead of just Kib/s could be acceptable.

Yet another informative post

Oh, do I love myself when I make these intelligent articles that actually inform people rather than annoy them. Well, turns out they are not such a common occurence on this blog, but hey, they’re usually longer. So, again, I’m not asking anything here, unlike the title seems to indicate, I’m actually teaching you something. Now you could go ahead and read everything about it in your favorite encyclopedia, cough, Wikipedia, cough, but sometimes doing that yourself and decifering the amalgamy of information to just what you want to know can be daunting. If you haven’t, take a look at my PS3 Sound Article, it’s all stuff straight out from encyclopedias and very technical articles, but reformulated just so the average person can understand it without having to do all the nasty research. So, let’s dive in!

Why should I know this?

The big question, why should you, of all people, care about kb, kB and KiB. What is it and what does it have to do with my life. Turns out, if you use a computer, it concerns you. If you use a cell phone with any form of data on it, it concerns you. If you pay an Internet bill, it concerns you. If you use an MP3 player (ie. an iPod), it concerns you. And if you’re reading this blog, this is sure to come in handy. Even if you’re pretty well versed in computing, there may be some things that’ll surprise you, so read on.

First off, the bit and the byte

In computers, data, like your Microsoft Office Word documents, pictures, videos and music, are all consisting of 1 and 0 somewhere in the memory of your computer (ie. your hard disk drive). This is called a binary system. Every electronic machine capable of computing something relies on a binary system to do it. Yup, from your digital watch to your cell phone, from your calculator to your oven’s dash (if it’s digital), and of course your computer, they all use a binary system.

The reason is pretty simple. The only way for electricty to be represented into computable data is to use its single invariably changing property, whether it’s on or off. This is represented by a 1 for on, and a 0 for off. And then, will all sorts of complicated mechanisms not explained here, cabling, chipsets and boards, along with billions of ones and zeros, comes to life everything digital you use in your daily life, like your brand new HDTV or your gaming console.

Let’s go back to computer data. Earlier we said it was all 1 and 0. Well, each time you have either a 1 or a 0, you have a bit. A single bit of data can have two possible meanings, 1 or 0. On a CD, this is represented by a hole (1) and no hole (0). This is why we call it “burning a CD”, because we literaly burn in a hole every time we get a bit representing a 1. And by combining all those bits together, we get a CD with stuff on it, like Music.

However, a bit alone cannot represent much, this is why in computing, most bits are coupled in series (technically a word, the smallest addressable sub-field, or useable piece of data, of a computer). The most common serie is one of 8 bit. For example, the 8 bit ASCII system used to represent raw text in its Western flavor, the letter a is represented by 01100001, which is a binary sequence of 8 bit, because it has a combination of 8 ones and zeros.

A single bit is represented by the lower case letter b.

Eventually however, counting data strictly in bits becomes tiresome. Only two letters is 16 bits, and 12 bits of text equals one letter and a half, which is impossible. This means that in an 8 bit system, counting the bits is useless, since you can’t really split them, they have to come in pairs of 8, always. And so, some brilliant guy in the past invented the byte. The byte is simply a measure of a single bit serie. Thus, in an 8 bit system, 1 byte is equal to 8 bit, so 2 letters is equal to 2 byte, a bit more representative. Additionally, the byte prevents impossible spliting since you cannot split 1 without getting a floating point number (ie. 0.5), which doesn’t exist in the binary system.

A single byte is represented by the upper case letter B.

A bit of history

Historically, a byte could represent any serie of bits, as long it matched its system’s base. If data was represented by a 4 bit system, a byte would equal 4 bits, if data was represented by a 16 bit system, a byte would equal 16 bits, and on and on. However, due to IBM’s System/360’s 8 bit architecture in the 1960s and the explosion in popularity of microcomputers based on 8 bit microprocessors in the 1980s, 8 bit remained the standard by which we measure data. Other things, like the more recent microcomputers, also know as your common Desktop PC, and microprocessors, also known today simply as processors or CPUs (ie. the Intel Pentium), have however migrated to a higher number of bits per bit serie for higher efficiency and bigger memory. You’ll probably hear a lot about that in the coming years with the mainstreaming of 64 bit microprocessor architectures to replace 32 bit systems, mainly due to the memory limitations of 32 bit architectures.

In other words, even though file systems (the thing that handles your files) are today reaching 128 bit architectures, data is still represented in an 8 bit format, where as a single byte represents 8 bit.

Handling larger numbers

Once computer inovation started to ramp up, the quantity of data we work with continually expanded. Before engineers knew it, they had to deal with thousands of bytes, instigating the need for yet another way to sum up data into smaller numbers. Unlike money, which exists since way longer than bits and bytes, it was easy to perceive how we would eventually reach much more than billions of bytes of data. To sum up things better, engineers decided to borrow on the SI system, otherwise known as the “Le Système International d’Unités”, or International System of Units, a French invention to sum up large numbers in the metric system. If you live elsewhere than in the United States, you’re already familiar with this notation appearing everything.

In the SI, a lower case krepresents a thousand, a capital M represents a million, a capital G represents a billion, etc. Each measure has its full written form too, which consists of a prefix to add to any measure. k equals kilo, so 1000 meters equals 1 kilometer (also km). 1000 grams equals 1 kilogram (also kg). Since the Earth’s circumference is roughly only 40,000 km, some measures like the megameter (Mm) have yet to see any real use. 40 megameter seems less impressive than 40,000 km.

The importance of case sensitivity: The SI system makes great use of lower and capital case for its acronym. The best way to show this is with an example. A million, or a mega, is represented by an upper case M. A megameter is thus represented as follows: Mm. However, a very close measure, written mm, is the millimeter. A milli, or 1/1000th, is represented by a lower case m. And so, very subtle case sensitivity like the lower case k of a thousand aren’t so unimportant anymore. You wouldn’t want to use a capital K because it means degree kelvin. This is often overlooked in the computer world however, where subdivions of a unit are impossible. Indeed, a millibyte or a millibit does not exist.

In the case of a bit, the SI system does well. Since the SI system is always used at the power of 10, it is possible to use it perfectly standardly with the bit. 1000 bits is equal to 1 kilobit (kb).

However, when it comes to bytes, the story was a bit different. Because of the computer’s binary nature, memory addresses (where the data is physically located on a memory chip) are written in binary sequences. As such, the number of addressable memory locations are counted with a power of 2. For example, an 8 bit address space has a total addressable memory of 256 bytes. However, newer 16 bit memory architectures at the time could now jump-start the 256 bytes limit produce a much more potent 65,536 bytes of addressing space. Applying the SI system at the power of 10 on this produced a somewhat awkward number, 65.536 kilobytes (kB). To remedy the situation, engineers took the SI system and made it at the power of 2 to match memory addressable space, so that a kilobyte would equal 1024 bytes while a kilobit still equaled 1000 bits, bringing confusion still reigning into the world of computing today and a nice round 64 kilobytes of memory for the 16 bit architecture.

Controversy

Invented in America where the metric system has not yet been adopted as of 2009, the secluded world of computing did not wake anyone up with their awkward borrow on the SI system until 1995, when the IUPAC and the NIST proposed to a newer unambiguous system to the IEC, only to be accepted in 1999, giving birth the kibibyte and the mebibyte, a play on the word kilobinary and megabinary, later followed by the other higher multiples in 2000.

However, by that time, it was already 40 years since the computer industry had been using the SI multiples at power of 2. 10 years after the standardization of the IEC format, adoption has been nearly innexistant, with the only systems using the measure being rare very recent Linux distributions. Even the latest Windows and Mac OS X still use the SI prefixes for bytes.

But using the SI prefixes for bytes isn’t necessarily wrong, since the IEC standard allows it. Only, you have to use them at the power of 10, so one kilobyte equals 1000 bytes, and one kibibyte equals 1024 bytes. It goes without saying that this is a highly contested and controversial standard. It does bring clarity and non ambiguousness, but it also brings confusion for legacy systems.

For example, MP3 player users are used to see their storage capacity in GB when it should be in GiB. Adding the i will obviously bring in a lot of questionning in stores where a brand might not use the same notation. Most store clerks would probably been unable to even explain the difference. This is why marketing forces at Apple and Microsoft, along with almost every hardware manufacturer, decided to keep the original measurement.

Additionally, there’s very little explanation to why this system should be implemented at all. Afterall, in traditionnal computing, a kilobyte never meant anything else than 1024 bytes, so there’s very little proof at how the IEC’s newer system may improve the situation. Memory manufacturers won’t start making memory systems in multiples of 10 and so many think that the confusing would just be transfered over to the kibibyte, which would often been mistook for a thousand bytes.

Adoption

However, when it comes to the educated world, standards make a long way. Most of today’s technical documentation and teaching material has adopted IEC’s standard, refering to kilobytes as multiples of 10, and kibibytes as multiples of 2. This means that computer classes in school will start teaching it that way, and that eventually, operating system makers and memory companies will have to adapt, regardless of controversy.

Conclusion

While your computer probably still uses the wrong notation, assuming you’re reading this article as of 2009, future systems will probably adopt the IEC standard, so let’s review the whole thing. I’ve also included the bit version of the IEC, known as the kibibit (Kib) and its brothers, in this review, along with an in-depth explanation of capitalization rules in red.

Note on k/K: Although the SI system makes an absolute use of a lower case k for a thousand because the upper case K is for a degree kelvin,  remember that the computer use of the SI system has never been standardized into the SI’s base or derived units and many sources suggest that a capital K can also be used for a thousand.

A KB or a Kb cannot be mistaken for a kelvinbyte or kelvinbit because it doesn’t make sense. Also, since the binary system doesn’t make use of subdivisions and thus does not possess the lower case d, c, m and other prefixes, only the k being lower case would make for an inconsistent notation.

For these reasons, the IEC chose to have the kibibyte with a capital K and many often use a capital K for a kilobyte and a kylobit. Note that the use of a lower case k (ie. kiB) for a kibibyte is not accepted.

As the writer of this article though, I am a very purist person when it comes to standards. Since the kilobyte and the kilobit are both borrowing on the SI standard, I think they should use a lower case k, regardless of application. However this hasn’t been standardized, and you can use whichever you think is better. The review here under uses a lower case k.

1 bit      (b)   = 1 bit          (b)
1 byte     (B)   = 8 bit          (b)

1 kilobyte (kB)  = 1000 bytes     (B)
1 megabyte (MB)  = 1000 kilobytes (kB)
1 gigabyte (GB)  = 1000 megabytes (MB)
1 terabyte (TB)  = 1000 gigabytes (GB)

1 kibibyte (KiB) = 1024 bytes     (B)
1 mebibyte (MiB) = 1024 kibibytes (KiB)
1 gibibyte (GiB) = 1024 mebibytes (MiB)
1 tebibyte (TiB) = 1024 gibibytes (GiB)

Higher prefixes can be seen here for the SI system, and here for the IEC system.

Kibibits, mebibits and else also exist, effectively meaning bit multiples at the power of 2, exactly like kibibytes, mebibytes and company. However, the IEC pretty much created this standard just for the sake of it, as it isn’t really useful. Maybe in the future 1024 bit architectures will be called 1 kibibit architectures, but most architectures are far from 1024 bit in any cases.

Also, another notation exists for bits instead of b. Literaly using the full word bit instead, however invariable. The SI prefix is the traditional k, M, G, etc. and the IEC prefix is Ki, Mi, Gi, etc. So this goes like this: Kibit, Mibit, Gibit. The IEC seems to particularily encourage this notation to further distinguish between a byte and a bit traditionnally only being difference by a lower and an upper case.

Ah… those new ads from Microsoft, how wonderfully full of bullshit. I’m sorry, but no, you won’t get what I actually call a laptop for under 1000$. Simply, no. It’s already hard enough to get quality from a 1000$ Desktop Machine, imagine a laptop.

So even though I hate Apple ads, their recent ones that attack Microsoft’s new ads are really great. They really mark a point on a certain level. No, I’m not 100% in accordance with what they say, but at least it’s less bullshit than Microsoft’s ads.

But I always think, Mac Geniuses… wish we had that in Canada. I think we do have an Apple store actually, yes, one! Five hour drive from where I live though, you have to be in Toronto to access it, bummer.

Some time, brilliant stuff happens on the Internet. What would you do without Charlie Bit my Finger on YouTube… Well, no, really smart stuff, like Wikipedia or Google.

In my opinion, there is now a third product on the Internet that I classify “as smart” as Google or Wikipedia, Wolfram Alpha. Not a very markety name, but hey, every geek knows Wolfram. Go there now, http://www.wolframalpha.com/ and input some mathematics, or anything. Take a look at the examples especially, cause at first you might think you’re just too stupid to use it.

Well, I’m thinking this might just make math homework a lot easier.

So, I work right at the side of some translators at work and I can’t get enough of those feminization discussions. What do you do if you have both genders being addressed by a pronoun in French, usually you use “ils”. So:

- Ils étaient allés voir leur grand-mère.

“Ils” here could mean two things: only men or both men and women.
The big fight with equality is trying to go against this concept, where “Elle” would only be women and not both, by making some weird things like this:

- Ils et elles étaient allé(e)s voir leur grand-mère.

While not technically wrong in any way, I’ve seen worse, and it’s far from pretty. So, I say, while we’re on the French gender mania, which makes it a very difficult language, why not push it further and include a neutral gender, which includes both.
For example:

- Ols étaient alléx voir leur grand-mère.

“Ol” or “Ols” in the plural form would be the neutral way of “they” in French. And, instead of putting an S or an E or an ES as a feminin/masuclin singular or plural form, a simple X would define both. The X is, like the feminin E or the S or the ES are mute, and Ol is pronounced with the O sound in French.

Making the step up from shared hosting is not always easy. We’ve recently done so with AMV-Canada.com. The forum was starting to get really slow, so we had to move on. We upgraded from 1and1 home shared hosting to Media Temple’s Grid-Service, a clustered hosting solution. The transition was smooth and we didn’t have a lot of new IT stuff to learn, since it really is just like an upgraded version of shared hosting. At least, it’s better than WebHostingBuzz Business web hosting, which royaly sucks.

But right now, as I write this, this blog is being hosted on that Grid-Service from Media Temple and the uptime hasn’t been so great. Despite being very small issues, we didn’t experience more than an hour of downtime, the topic of the issues is what’s worrying. For example, we’ve been out of access of the control panel, our only way of managing our server, for about 6 hours and we’ve had issues where for about an hour, our MySQL server was out of memory. Despite it being slow, that kind of thing never happened on 1and1’s shared hosting service, kind of ironic.

Even more worrying, here’s what we’ve had like response from Media Temple support:

Our Engineers responded to reports regarding ”too many connections” on one of the database servers. The issue was caused by high usage on the server, and our engineers were quick to reduce the load. The issue has now been resolved and connectivity to the database has been restored.

Quick? One hour? Too many connections on one of the database server? Wait, I thought I was on a clustered service built to handle that kind of scaling issue. What’s more, my usage of the server was close to nothing, nobody was on the forum that morning. We’ve asked around and nobody noticed the downtime except us both admins. I don’t know for you, but to know that having a major traffic increase on my database-powered site would bring it down, and a couple of other clients too, for an hour, I’d put in question the efficiency of this clustered service. It’s the time it takes to resolve the same issue with a dedicated solution. Even cheap VPS solutions are still isolated server instances where if you crash the server, you don’t crash others. It also means if somebody else crashes the server, they don’t crash yours, only their instance.

Again, more worries, some other clusters got affected really bad by storage issues, seeing downtime for as much as 96 hours. That kind of thing can be avoided by monitoring the hardware first-hand. Some people do this right though, like Rackspace. Their clustered hosting solution, called Cloud Sites (aka Mosso), doesn’t suffer from the same major issues. But it does suffer from the all the intermitent database, server and control panel things, although, considerably less.

Dedicated solutions from Media Temple, however, including their VPS, have and are still known for their rock solid stability and hardware. Never will you see anything else than a really occasional maintenance window on VPS systems and even less on truly dedicated, same goes for Rackspace. So, is the clustered architecture at fault?

In all regards, it seems to be. It goes without saying that clustered hosting is as problematic as shared hosting, if not more, but it’s faster, especially Rackspace. I’m not sure it’s the kind of service you’d want to have for a mission critical application though, such as having a site available during a special event, say, Engadget updating an Apple conference in real-time.

Another problem with non-dedicated services is that you have no control on the software maintenance windows, much less hardware. What if they decide, depsite saying so a week ahead, to update some software on your cluster during your money-making event, bringing intermitent availability right in it. There’s no reason why it shouldn’t happen. I often thought clustered meant that whatever happened, our site would just be seemlessly tranfered over to another node and not have to suffer any downtime because of hardware or software maintenance. But truthfuly, you only get that kind of thing on custom dedicated solutions that use truly resilient clustering solutions, like Google does. As the tradition goes: – If Google is down, it’s probably because your own connection is down, not Google’s.

With dedicated solutions or any kind of VPS though, you retain complete control over the timeframe to do software maintenance (and hardware if you have a truly dedicated solution). That way, you can decide to check for problems and do maintenance before or after the event, not right in the middle of it.

So think about it, is the IT effort of managing your own dedicated solution, or a pricier managed solution, worthier than not having control over the timeframe to do maintenance. This should be a decision-making factor for your next hosting solution. It doesn’t mean that clustered services aren’t right for you, it just means that if you have a real-time/real-life event-dependent web site like Engadget, dedicated services might be a smarter option.

Media Temple Status Page
Rackspace Mosso Status Page