The IT Business

Interesting concept: Ksplice: Rebootless Linux kernel security updates. As long a patch doesn´t change data structures in the kernel, ksplice can update a running kernel.

Ksplice allows system administrators to apply security patches to the Linux kernel without having to reboot. Ksplice takes as input a source code change in unified diff format and the kernel source code to be patched, and it applies the patch to the corresponding running kernel. The running kernel does not need to have been prepared in advance in any way.

This is really interesting. I´ve talked about this effect at my presentation on Friday. There is an effect in statistics called pseudo correlation. You have an pseudo correllation, when you can proof a correlation between two attributes , albeit the real relation lies somewhere else.

I thought a little bit about virtualisation this weekend after some interesting comments of a customer about his experiences with virtualisation.

Are hypervisor based virtualisation mechanisms really the way to go? Think about the timing problems of VMware. The problems to virtualize dozens of timer ticks,eating away a good amount of the capacity of your system. Or the problems with I/O intensive tasks. You can throw more CPUs to the problem.But: I´m tempted to think, that solutions like branded zones are a more viable solution to virtualize for unix systems. You stay with a single kernel infrastructure mapped to different target operating systems by special mechanisms ... like the branded zones, perhaps in conjunction with resource management. Looks more efficient to me than simulating the complete system.

This quote found at Dave Edstrom´s blog made my day ...

"There are two types of users - those who are ready to spend a lot of time in order to save money, and those who are ready to spend a lot of money in order to save time.”
- Mårten Mickos, CEO MySQL

My brother is consultant for increasing the energy efficency of buildings. Not long ago i had an interesting discussion with him: In IT we use the example of the electricity grid and a central power generating plants as the future of a utility based computing. The problem: My brother told me, that this is a model of the past. There is a massive trend in the generation of electricity away from the big central plant, the future is the small decentralized power plant to generate the electric power near the consumers of power to get rid of the transport losses (those losses accumulate to an primary energy factor of 1:3. For each consumed kilwatt, you have to put 3 into the power grid).

There are even thought games about virtual power plants by generating electical power directly at the households. All this micro plants are operated and controlled by a central control. When you need more energy in a region, the microplants in the region start to produce more energy to feed the energy in the regional grid.

Perhaps this is the way to go in utility computing. The systems are at the consumer premises, operated and managed by a central . When other customers has a surge in it´s need for more compute power , the compute power is generated by the micro-datacenters at the others sites, for example at sites, where the customer can´t load it´s compute system (a good argument for cryptography everywhere). In this case you have a fast and low-latency access to your compute power at normal operation but access to a vast amount addtional capacity. Smaller customers in a region could even start without own machines and when they demand more than a certain amount of capacity, and own micro compute plant will be installed at this site.

Sometimes it´s really easy to see, if a writer is partisan. Okay, i´m pro-sun ... that´s obvious and i´m admitting that. Hey, i´m working for Sun. But i saw a really nice example for an partisan "independent" writer. He criticize some design decisions at UltraSPARC T2+over a length of two paragraphs. That´s okay, but it throws an interesting light, when you celebreate a watercooled IBM p575 in a nonstandard formfactor of 23" width as the way to go in computing at the same time.

But, hey ... it was pretty obvious that TPM is heavly IBM biased before as i´m biased for Sun. But at least I don´t pretend to be independent

It´s nice that he references to ZFS in his article, but i think his opinion stated in #"PostgreSQL to Scale to 1 Biilllliooonnnn Users, Dr Evil would be proud" is dangerous:

Backup is irrelevant for those of you who care about this discussion. LVM/ZFS snapshots are the rule of the land.

Well, when the first disaster hits your datacenter (and such disaster begin with a failed disk too much) , you will recognize that you are toast without tapes or at least with a disk-based backup on a different storage as a last line of defence. Neverever rely on a single pool of disk blocks on a single machine for disaster recovery, even when it gives you features like snapshot to freeze the disk state at one or multiple points in time.

One observation with virtual machines on my Macbook Pro:

# ntpdate 130.149.17.21
30 Mar 16:58:00 ntpdate[710]: step time server 130.149.17.21 offset 82.381455 sec
# ntpdate 130.149.17.21
30 Mar 16:59:07 ntpdate[711]: step time server 130.149.17.21 offset 33.485510 sec