Apache finally signs off Hadoop database... after 7 years of development A completed first version of what’s been called the Hadoop Database has been released, seven years after work started. HBase 1.0 was signed off on Tuesday by the Apache Software Foundation. The system has been built to work with different implementations of Hadoop. Michael Stack, vice president of ASF HBase, welcomed the news …
You can probably run HBase itself on anything that can run Java 6 - its dirty little secret is that almost all of its work is done on-heap inside the JVM. Unfortunately the rest of the Hadoop stack (upon which you should probably be running HBase) relies heavily on native extensions, written for x86, to achieve the performance that it advertises. Essential things like compression will run like an absolute dog on anything but x86 Linux.
No,
The only 'extension' to Hadoop / HBase is the use of encryption that came from Intel. Compression is either Java or native. So if you compile the native compression you can link to it.
You can run HBase out of the box on any Linux X86 platform which would include AMD chips.
If you want, you could run it on Hadoop that runs on Windows, but only Hortonworks would probably support that. (Microsoft funded that project.)
Hadoop hasn't been ported to other OSs although in theory you could port it to run on the mainframe in their Linux distro...
But really, who's running PowerPC or Sun boxes these days that isn't running a Linux variant?
Read my post very carefully Ian. It agrees in its entirety with yours.
Though you are mistaken on one other key point - in addition to accelerated encryption (you can do HDFS encryption in Java without acceleration should you choose [don't]), the native extensions are required for short-circuit reads and centralized cache management. These are two of the key features that take hadoop from the slow, clunky batch land of 2009 to the lightning-fast useful land of today. Even if you've got hadoop.dll/hdfs.dll in place of libhadoop.so/libhdfs.so, you still can't do everything - notably the short circuit reads.
So, to reiterate, yes you *could* run it on ARM, and it has been done. HBase is particularly easy to run on other platforms because it's as-good-as 100% Java (with all the pain/pleasure that brings). Unfortunately,t he further and further you get from intel x86 and linux, the slower and uglier it's going to get, so while you could do it, don't.
Speed is relative.
And yes, several things got dumped in to HBase which should have core components rewritten.
I was saying that HBase will run on a lot of platforms and even without the native Intel additions, its fairly fast if you know what you're doing.
W.R.T encryption... and security... I'm unfortunately not allowed to talk about that. Maybe in a couple of months... ;-)
My point is that outside of Linux, only Windows is a 'viable' platform thanks to Hortonworks working with Microsoft. And while that's possible, its not recommended. No Power PC or Sparc chipped versions exist although they could. Even on the mainframe. (Which can run a Linux partition (LPAR) )
I've had customers move away from HBase because it wasn't stable (Still isn't) and you really need to have an admin who knows both Hadoop/HBase but also is a software developer and knows distributed systems. (That's a lot of things to ask and besides myself... maybe a handfull of people can do it. )
Jonathan Gray comes to mind.
The real trouble of HBase is that people don't understand what it is, how to use it and why you shouldn't consider it an RDBMS (Sorry Splice Machine guys...) ;-)
And yes, we're in violent agreement. ;-P
I have to ask.. you in the UK? If so, I'm pretty sure I know who you are, even though we haven't met.
(And if you're in Germany, the odds are we have met. ;-)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
