This is one of the things that I think the Atlantica team really did well on. Because they have an "alternate Earth" as a background, they were allowed to present different cultures with different art styles.
The biggest problems we've seen from a monitoring perspective is that most systems really do have a hard time scaling to large levels and being usable. [A common trick (and one we employ) is to have a multi-tier monitoring system in place, where one monitoring stack monitors the monitoring stack that is actually watching the service/hosts.]
Once one gets past that hurdle, the tricky part is dealing with the "it is OK if X% of my machines are down". Most monitoring systems that I've dealt with are based around the view that they are monitoring a single host/single service and not a collection of hosts where it is OK if chunks disappear. For those types of problems, one still ends up writing a lot of custom smarts it seems.
Who wants a turn-based RPG anymore? It's all about the flashy graphics and real-time combat.
That's why it is interesting watching to see how successful Atlantica Online is going to be (or already is?). A turn-based MMORPG sounds horrible, but they've really done a great job at game mechanics. The irony is that they likely had to do F2P because of the impressions that turn-based gives, but they are probably making money hands-over-fist with the micropayment structure they've setup. It isn't usual to hear of people who have spent hundreds if not thousands of dollars on it.
None of the articles say it, but they are probably talking about HBase. If this is the case, this is seriously old news.
HBase was started by the Powerset guys before being acquired by Microsoft. After the acquisition there was a lot of concern in the Hadoop community about whether the Powerset guys would be allowed to continue to contribute. They have, and as far as I can tell, the community is not particularly concerned about MS's involvement.
Kerberized NFS support is required for full NFSv4 compliance. You might be surprised to know that Linux, AIX, and NetApp (although their Kerberos support is... lacking) all have the capability to do it. OS X has rudimentary support for v4, haven't tested its krb5 compliance.
Additionally, most stacks either out there or in development add Kerberized NFS support to v3 while they are adding v4.
On some operating systems (Solaris comes to mind), the crash dump is stored in swap compressed, so 2x may still be overkill even if you are doing crash dump testing.
The largest "real, in-use" Hadoop cluster that Yahoo! has is around 2000 nodes, counting a dedicated name node. As far as we're aware, we've got the largest Hadoop cluster. [If there is a bigger one, we'd love to talk to you and compare notes.:) ]
That said, we do have Hadoop running on tens of thousands of machines. Just not as one big cluster.
It is also worth pointing out, that most of our clusters are multi-user, multi-application. The number of nodes is really more indicative of the size of the Hadoop distributed file system than the number of nodes given to a particular application in our (grid team's) use case.
I find it interesting in the 'last decade' sort of problems: root drives, RDBMS, etc. But unless it is linked with something like Lustre, I'm not sure if it is ready to take on the problems of this decade and Big Storage a la Google File System or Hadoop Distributed File System. Even a 70TB FS of the size you mentioned are starting to look tiny in comparison.
[For the record, I've used both ZFS and HDFS. Our largest HDFS--and we have multiple--has around 3-4PB, counting the space required by the 1:3 replication factor we use.]
"The OpenSolaris kernel was a welcomed surprise: it was the only body of source code that did not require any extensions to CScout in order to compile."
Given that the Solaris kernel has been compiled by two very different compilers (Sun Studio, of course, and gcc), it isn't that surprising. Because of the compiler issues, it is likely the most ANSI compliant of the bunch.
I have this image of millions of angry lines of code rampaging across the steppings, raping and pillaging peaceful programs, stealing their output and burning their execution platforms. A bit like badly written map/reduce code...
An alternate universe where not only is the date weird (maybe the leap day happens in January?) but Microsoft employees apparently don't know their CEO is Steve Ballmer and need to have this explained in internal documents on a regular basis.
Google hasn't released anything other than papers on GFS and their implementation of MapReduce. At this point, though, I'm not sure it matters since we have Hadoop (which, being mainly Java, C, and a little bash) runs perfectly fine on all of the major operating systems, including Windows.
So, Hypertable runs on top of Hadoop. We don't use Hypertable (or HBase) so I can't commen on those. I can share some of our experiences with Hadoop though. I think it is safe to say that it scales quite well for the vast majority of people who need it. Let's deep dive for a bit...
Hadoop keeps all of its file system metadata in memory on a machine called the name node. This includes information about block placement and which files are allocated which blocks. Therefore, the big crunch we've seen is the total amount of memory available to the JVM's heap. With a 16G machine (with ~14.5G heap) for the name node and ~2000 machines acting as data nodes, we're scaling to somewhere between 12-18 million or so files [it's been a while since I've looked...:) ]. Our capacity should be around 5PB or so, but keep in mind that large chunks of that are taken up by block replication and that, for the most part, that number is tied into the physical size of the drives in use.... and, yes, this is all on commodity machines that you can get from pretty much anyone.
We're working on making it scale better, of course. But we've come a long way in a really short time. [We've doubled capacity in less than... six months? Something like that.]
It is a bit disappointing and surprising to see that the author spent more time talking about Silverlight and moderation and other things when he could have written about the *other* open source things that Yahoo! either supports directly or where patches have been contributed. Where is the discussion about Hadoop, Pig, ZooKeeper, YUI, PHP,... ? Do these not impact Linux users from his point of view?
Having Sun rename a product is probably one of the worst things that could happen. I waiting for the launches of Sun Java System Database Enterprise Server and Sun Java System Database Open Source Server to happen any day now.
Computer versions of Solitaire pre-date the Windows versions. For example, Compute! ran listings for Canfield in one of its last issues that had type-it-in-yourself code (1988). Other links on the the Compute! index page show even earlier versions (e.g.,1986).
My parents live where they can't get cable (very rural) and they currently do not subscribe to satellite. I was actually surprised to see ads on TV talking about where to log in online or call to get the coupons "starting January 2, 2008". Since this was a week or two ago, it is pretty clear they are trying to get the word out and trying to get it out early.
Slashdot Mirror
Those jobs do exist. You just have to know where to find them. :)
Unless there is a mark that says "Item was bought", I'd argue that most players wouldn't know the difference between a bought or won item anyway.
This is one of the things that I think the Atlantica team really did well on. Because they have an "alternate Earth" as a background, they were allowed to present different cultures with different art styles.
The biggest problems we've seen from a monitoring perspective is that most systems really do have a hard time scaling to large levels and being usable. [A common trick (and one we employ) is to have a multi-tier monitoring system in place, where one monitoring stack monitors the monitoring stack that is actually watching the service/hosts.]
Once one gets past that hurdle, the tricky part is dealing with the "it is OK if X% of my machines are down". Most monitoring systems that I've dealt with are based around the view that they are monitoring a single host/single service and not a collection of hosts where it is OK if chunks disappear. For those types of problems, one still ends up writing a lot of custom smarts it seems.
I'm curious to know how you define a 'corporate farm'. I really hope you don't think that every Xxxx Farms, Inc. is magically part of ADM.
I see Hadoop Summit door prizes.
That's why it is interesting watching to see how successful Atlantica Online is going to be (or already is?). A turn-based MMORPG sounds horrible, but they've really done a great job at game mechanics. The irony is that they likely had to do F2P because of the impressions that turn-based gives, but they are probably making money hands-over-fist with the micropayment structure they've setup. It isn't usual to hear of people who have spent hundreds if not thousands of dollars on it.
None of the articles say it, but they are probably talking about HBase. If this is the case, this is seriously old news.
HBase was started by the Powerset guys before being acquired by Microsoft. After the acquisition there was a lot of concern in the Hadoop community about whether the Powerset guys would be allowed to continue to contribute. They have, and as far as I can tell, the community is not particularly concerned about MS's involvement.
Kerberized NFS support is required for full NFSv4 compliance. You might be surprised to know that Linux, AIX, and NetApp (although their Kerberos support is ... lacking) all have the capability to do it. OS X has rudimentary support for v4, haven't tested its krb5 compliance.
Additionally, most stacks either out there or in development add Kerberized NFS support to v3 while they are adding v4.
Correct. Solaris does not support memory overcommit. Linux does, not sure about others.
On some operating systems (Solaris comes to mind), the crash dump is stored in swap compressed, so 2x may still be overkill even if you are doing crash dump testing.
No, you use a few thousand cheap x86 boxes to decentralize your storage and move your computations to where the data is located. :)
Actually...
The largest "real, in-use" Hadoop cluster that Yahoo! has is around 2000 nodes, counting a dedicated name node. As far as we're aware, we've got the largest Hadoop cluster. [If there is a bigger one, we'd love to talk to you and compare notes. :) ]
That said, we do have Hadoop running on tens of thousands of machines. Just not as one big cluster.
It is also worth pointing out, that most of our clusters are multi-user, multi-application. The number of nodes is really more indicative of the size of the Hadoop distributed file system than the number of nodes given to a particular application in our (grid team's) use case.
There is a lot more about Hadoop, and Yahoo!'s particular Hadoop usage for internal utility-type computing, at http://wiki.apache.org/hadoop/HadoopPresentations .
Bonus points if you can get paid by the company to work on open source.
[This can work for non-coders too. It is an absolute joy to be able to go to USENIX or ApacheCon or whatever and discuss/share common problems. ]
I find it interesting in the 'last decade' sort of problems: root drives, RDBMS, etc. But unless it is linked with something like Lustre, I'm not sure if it is ready to take on the problems of this decade and Big Storage a la Google File System or Hadoop Distributed File System. Even a 70TB FS of the size you mentioned are starting to look tiny in comparison.
[For the record, I've used both ZFS and HDFS. Our largest HDFS--and we have multiple--has around 3-4PB, counting the space required by the 1:3 replication factor we use.]
ZFS was good enough for me.
"The OpenSolaris kernel was a welcomed surprise: it was the only body of source code that did not require any extensions to CScout in order to compile."
Given that the Solaris kernel has been compiled by two very different compilers (Sun Studio, of course, and gcc), it isn't that surprising. Because of the compiler issues, it is likely the most ANSI compliant of the bunch.
An alternate universe where not only is the date weird (maybe the leap day happens in January?) but Microsoft employees apparently don't know their CEO is Steve Ballmer and need to have this explained in internal documents on a regular basis.
Google hasn't released anything other than papers on GFS and their implementation of MapReduce. At this point, though, I'm not sure it matters since we have Hadoop (which, being mainly Java, C, and a little bash) runs perfectly fine on all of the major operating systems, including Windows.
So, Hypertable runs on top of Hadoop. We don't use Hypertable (or HBase) so I can't commen on those. I can share some of our experiences with Hadoop though. I think it is safe to say that it scales quite well for the vast majority of people who need it. Let's deep dive for a bit...
:) ]. Our capacity should be around 5PB or so, but keep in mind that large chunks of that are taken up by block replication and that, for the most part, that number is tied into the physical size of the drives in use. ... and, yes, this is all on commodity machines that you can get from pretty much anyone.
Hadoop keeps all of its file system metadata in memory on a machine called the name node. This includes information about block placement and which files are allocated which blocks. Therefore, the big crunch we've seen is the total amount of memory available to the JVM's heap. With a 16G machine (with ~14.5G heap) for the name node and ~2000 machines acting as data nodes, we're scaling to somewhere between 12-18 million or so files [it's been a while since I've looked...
We're working on making it scale better, of course. But we've come a long way in a really short time. [We've doubled capacity in less than... six months? Something like that.]
It is a bit disappointing and surprising to see that the author spent more time talking about Silverlight and moderation and other things when he could have written about the *other* open source things that Yahoo! either supports directly or where patches have been contributed. Where is the discussion about Hadoop, Pig, ZooKeeper, YUI, PHP, ... ? Do these not impact Linux users from his point of view?
Having Sun rename a product is probably one of the worst things that could happen. I waiting for the launches of Sun Java System Database Enterprise Server and Sun Java System Database Open Source Server to happen any day now.
Computer versions of Solitaire pre-date the Windows versions. For example, Compute! ran listings for Canfield in one of its last issues that had type-it-in-yourself code (1988). Other links on the the Compute! index page show even earlier versions (e.g.,1986).
My parents live where they can't get cable (very rural) and they currently do not subscribe to satellite. I was actually surprised to see ads on TV talking about where to log in online or call to get the coupons "starting January 2, 2008". Since this was a week or two ago, it is pretty clear they are trying to get the word out and trying to get it out early.