I moved to the opposite direction, from disks to tapes. We are a small IT company with less than 10 employees, mostly developers, no dedicated sysadmin, but quite a few servers. We had no previous experiences, so our backup "methodolody" is only slowly improving in ad-hoc ways. First we had backups of critical data stored on online disks. Source code and a very few other things. No configuration files, no database. There were only ad-hoc copies of the latters. After a while I started to backup more data, and started to use a centralized backup software (namely Bacula). Backups were still written to online disks. Everything was online, and we are reguralry attacked by hackers. Not a good combination. In the next step I tried to occassionally copy the online backups to offline disks. That did not really worked. Copyying all backups were time consuming, and I usually forgot it or do not have time for it. Large hard disks are still not that cheap, and they cannot be simply taken out the server and put back. The drives must be mounted on a tray, which require additional costs and work. Most people forget that if I have a 300 GB dataset, then I need about 30 * 300 GB backup space. There are tricks to reduce that, but that makes everything more difficult, more time consuming, less safe. I started to use tapes 2 years ago, after I recognized that our backup software supports that better. I thought that tape for us will always be a luxury, but it makes things much simpler. Equipment cost was definitely not a motivation. And indeed that is what happened, now everything is super simple. Bacula tells me what tape should I put into the drive, or if I need to buy a new tape. We have offline copies and multiples copies recorded at different dates. We can retrieve and compare data from 1 day ago, 2 days, one week, one month one yeat, whaterver I want. That is a nice safety, much-much better than we had previously. Bacause tapes are cheaper, I do not mind adding new data to backup and new tapes, just to make things simple. The funny thing is that I believe we shortly reach the point when our very simple, very safe, tape based backup system will be actually cheaper than the equal hard disk based, even if I do not count labour, only equipment. I am sure that a larger organization with real sysadmins can do it better, but I am quite happy with our current state.

The submitter does NOT complain about Google's ability to catch spam! He asks why Gmail does not REJECT obvious spam. Rejecting an email means that - in this case the Gmail - server does not even accept it. In such cases the sender gets back a Delivery Status Notification from his own server, telling him that his email did not go through because of such and such error. An important point here is that the email is not lost without any notification. The sender can try to contact the recipient in another way. Actually this may be better than putting the email into a spam folder if that is not monitored regularly, or at all. Yes, this is a valid question, but almost none have undersood it.

I do not understand this complaint about unix configuration either. I am a relatively new Linux user (about 5 years, compared to 20 years on Windows), and I find the Unix configuration system is far-far better than the mess in Windows. Everything is in the /etc/ directory. In the rare cases when I install a tarball instead of a package, the configuration files are in /opt//conf. I manage about about 50 virtual servers, it really works well.

Text configuration files are easily managed by standard command line tools, including diffing and merging changes during upgrade, and non-interactive modifications.

Can you read at all? He is saying among others that frequently some data (NoSQL tuned in one way) is better than no data at all (ACID). Is it better if my doctor knows only half of my previous illnesses, or if none at all in emergency?

Demagogy. Your example has nothing to do with ACID. Such a case would inditate wrong data entry or a client software bug.

By the way, such errors do occur. Even in systems where the database is ACID.

I have not read the article, but I guess they store either very frequent data (measurements) in NoSQL, or large data (3D images). Depending on nonfunctional requirements, neither is possible at all, or cost-effective with RDBMS.

No, no, it is the Apacha Foundation.

I haven't wanted to hurt anybody's feelings. We also have four Ubuntu desktops and two Ubuntu laptops in our house, including both old and current high-end machines. However, I believe it is quite unusual when it turns out that 15% of some group of teenagers uses Linux, and it must have some rational reason. I am sure that there are other reasons too, like my son's advocacy, but having old PCs are likely a reason.

I believe you make my point stronger by showing that even with such small resources and incompetent developers, it was possible to create the most moddable game (ever?). Without actually putting any effort into moddability... That made it possible to spend all of their limited resources on a good game design. That is not a small feat.

I worked in both C and Java for many years, and regarding moddability, no, it is not possible to beat Java using C with equal resources. My son's code regurly replace builtin Minecraft classes, runtime, without a significant effort, if the existing extension system does not provides (yet) the necessary hooks for him.

I am a server side developer for 14 years on a single system. We use MySQL but still not its ACID table types. After so much time maybe I am in the position that I can state, that, no, most of our data does not require ACID. Even which would require it theoretically is doing fine after 14 years.

What we would indeed need, is the multi-datacenter capability. Which you get for free with Cassandra... We also sorely needed performance a few years ago (15k SAS drives was slow after an internet hiccup for example), but SSD drives helped in that. Again we could get infinite scalability with Cassandra for free.

You must choose in such a situation: either the - only theoretically needed - ACID, or the actually performing and highly available NoSQL with its additional operations, coding burden?

I only have a little experience with Cassandra, but I can tell you, that consistency is very easily tuneable in it and it is also provides durability. Atomicity is restricted (AFAIK you can get atomicity if all your data goes onto a single data partition). Isolation does not exist.

I believe that it is very easy to say that something need ACID, while actually most data does not require ACID. They can, and as I read the article they do use relational database for those data which actually require ACID. NoSQL does not mean that you are in a constant state of getting garbage from your database. And yes, RDBMS are not failsafe either, if nothing else there are coding errors in the applications which use them.

I would add that Java also made it trivial to run Minecraft on Linux. More than one of his friends play Minecraft on Ubuntu, I guess they have old PC-s.

I think one of the things which made Minecraft popular is Java. There are a huge number of plugins and mods, these wouldn't happen without Java. It is easy to reverse engineer and modify anything in Minecraft exactly because it is in Java. Even its plugin system was written by an external developer! I do not really know Minecraft myself, but my 13 old years son plays Minecraft, and he spent months coding Minecraft extensions, and as far as I hear from him, a usual server uses a very large number of extensions.

Java is not ideal for graphics intensive applications, but it is also not that bad either. Minecraft (without mods) does not represent what is possible in Java, becuase it is very under-optimized. The new 1.8 version is much faster, but there is still much room for optimization.

This is similar to why PHP web softwares are very popular, they are not perfect, but they are very easy to be modified.

That is a nice theory, but it is false. You can easily validate it yourself, as I did, see my other post about the results.

Huh? The two cores of the Bulldozer module indeed has one common FPU unit, but that is a 256 bit one, which can be divided into two 128 bit unit (or even into four 64 bit unit!). I did test FPU performance and in the worst case it was 25% slower, in the best case it was actually faster, when I run two threads on a single module vs on two modules. Usually the difference is very small. Please do not compare the AMD Bulldozer architecture to Intel Hyperthreading, the two technology has very different purposes.

I agree. I went from cheap to the most expensive consumer routers in several steps (the current one is above 200$) but I have trivial problems with all. Because of this I slowly migrate all of their functions to a PC. If the current router fails, I will not buy a new one. I believe these home routers are useful only if you do not want to touch them at all (after setting the password). If you do anything more, and have enough space, use a real PC. It does not consume too much power, a usual PC uses less than 80W if it is idle. If you do not have a surplus PC, thenn buy an AMD AM1 APU with an ITX or mATX motherboard, and the idle consumption will be less than 25W. And you can use your favourite unix/linux OS.