I have never been too impressed with what Ubuntu came witch, such as the Ubuntu netbook edition, although it did the job for the HP mini I had. Granted that I never tried Ubuntu Unity, but even so, I wanted to stick with Fedora and Gnome…

But then came Gnome3, the new version of this Linux desktop.
It is not that I dislike the new Gnome shell. It is very pretty, actually. The problem to me is that, although it is pretty, unlike, for example Apple products, the Gnome developers and designers could not bring two more factors in the equation : the intuitiveness and my desktop production style -which I am sure I share with quite a few people-. I am certainly no Apple fan, quite the contrary, actually, but I really missed here something not only beautiful and user-friendly, but also productive.

I need different fixed desktops for my music, to browse the web, to work on documents and to read my emails. This was simply not possible with Gnome, since the desktops automatically close when they are empty, changing their order. There is probably a way to fix that, but this was not the sole issue…

One more thing is that the 3D effects were far too slow to be usable. After having opened the fourth application, my desktop, that works like a charm under Gnome2 and KDE, became really too slow. Not an option for me.

The last thing that finished convincing me to switch to KDE was the wireless device that worked on the LiveCD but not after being installed. That really upset me, especially given that it works well under KDE (hence not a Linux kernel problem).

For these three reasons : because it is not sufficiently intuitive, because it is too slow, and because I did not want to lose 20 hours fixing a wifi stick that would work on another desktop, I switched to KDE.

I really think that the new Gnome shell is pretty and has value for some users, but not enough for me. That is something I love about Free Software : you don’t like what you have ? Then switch to something else ! Competition is definitely good for everybody…

[herve@netbook2 ~]$ df -h
Filesystem            Size  Used Avail Use% Mounted on
/dev/mapper/luks-6f51b932-7e77-4922-a484-ec2d25fbcff4
2.9G  2.4G  518M 83% /
tmpfs                 497M  456K  497M   1% /dev/shm
/dev/sda1             485M   64M  396M  14% /boot
/dev/mapper/luks-27f1ea87-e3f1-4b8e-9cc5-50c543d5601c
20G  5.3G   13G  29% /home
/dev/mapper/luks-3fe2459e-5b8f-4c28-ae14-8e234a66d774
4.9G  138M  4.5G   3% /opt
/dev/mapper/luks-7cae3953-7d7d-4faa-a684-232f050287f6
1006M  100M  856M  11% /tmp
/dev/mapper/luks-1f05825f-bc91-4a53-9194-7e6ae271fa59
15G  2.4G   12G  18% /usr

To increase the logical volume, I use the traditional lvextend

[root@netbook2 herve]# lvextend -L 4G /dev/mapper/vg_netbook-LogVol00
Extending logical volume LogVol00 to 4.00 GiB
Logical volume LogVol00 successfully resized

I see here that the (long) device mapper file is a soft link that points to a file that is much easier to use

[root@netbook2 herve]# ll /dev/mapper/luks-6f51b932-7e77-4922-a484-ec2d25fbcff4
lrwxrwxrwx. 1 root root 7 Mar  6 08:51 /dev/mapper/luks-6f51b932-7e77-4922-a484-ec2d25fbcff4 -> ../dm-3

The cryptsetup then resizes the encrypted logical volume

[root@netbook2 herve]# cryptsetup resize /dev/dm-3

And finally I can increase the filesystem

[root@netbook2 herve]# resize2fs /dev/dm-3
resize2fs 1.41.12 (17-May-2010)
Filesystem at /dev/dem-3 is mounted on /; on-line resizing required
old desc_blocks = 1, new_desc_blocks = 1
Performing an on-line resize of /dev/dm-3 to 1048071 (4k) blocks.
The filesystem on /dev/dm-3 is now 1048071 blocks long.

Checking if everything went well (in bold, the size has increased):

[root@netbook2 herve]# df -h
Filesystem            Size  Used Avail Use% Mounted on
/dev/mapper/luks-6f51b932-7e77-4922-a484-ec2d25fbcff4
4.0G  2.4G  1.6G 60% /
tmpfs                 497M  456K  497M   1% /dev/shm
/dev/sda1             485M   64M  396M  14% /boot
/dev/mapper/luks-27f1ea87-e3f1-4b8e-9cc5-50c543d5601c
20G  5.3G   13G  29% /home
/dev/mapper/luks-3fe2459e-5b8f-4c28-ae14-8e234a66d774
4.9G  138M  4.5G   3% /opt
/dev/mapper/luks-7cae3953-7d7d-4faa-a684-232f050287f6
1006M  100M  856M  11% /tmp
/dev/mapper/luks-1f05825f-bc91-4a53-9194-7e6ae271fa59
15G  2.4G   12G  18% /usr

This Google Summer of Code project written in Python allows to execute a SSH command, to copy files, to rsync directories and to kill processes (among others) on multiple hosts

The install

First download the .tar.gz document
# wget http://parallel-ssh.googlecode.com/files/pssh-2.2.2.tar.gz

untar it
# tar -xzvf pssh-2.2.2.tar.gz
# cd pssh-2.2.2

To install (you need Python, it worked on RHEL 5.3):
# python setup.py install

The configuration

Then to execute parallel commands, create a file with all IP/hostnames you
need (one per line) like this
# cat hosts.txt
192.168.10.1
192.168.10.2

Create the directories “stderr” and “stdout”
# mkdir /tmp/stderr
# mkdir /tmp/stdout

and that’s it !

How to use it ?

If you want to execute the command hostname ?
# pssh -A -e errors -o /tmp/stdout -e /tmp/stderr -h hosts.txt -l root hostname

-A will ask the root password inline
-o will put the standard output in one file per host in the directory /tmp/stdout
-e will put the standard errors in one file per host in the directory /tmp/stderr
-l is the user that will execute the command

For my benchmark I nedded to change the starting script run.sh, so I edited it on one virtual machine and then used pscp to copy it on all the nodes in the directory /root/java/
# pscp -A -h hosts.txt -l root /root/java/run.sh /root/java/

And finally I executed the benchmark on my 32 nodes
# pssh -A -e errors -o results -h hosts.txt -l root /root/java/run.sh

Have you forgotten one JVM garbage collection parameter ? No worries, pnuke will make sure that all java processes are stopped on all servers
# pnuke -A -h hosts.txt -l root java

During the last past years, the rise of Linux-based operating systems, mostly on x86 architectures, was spectacular, because distributions such as Red Hat were able to provide a lower-cost alternative in terms of acquisition. However, the vast majority of these decision-makers, who, besides, also embraced Linux in their datacenters, thinks that Unix-based operating systems still bring more value in the following categories:

- availibility

Indeed, when it comes to availability, Unices have an advantage. The maturity of the product is higher, but above all, the integration of the platform to the operating system and to the application give an indisputable advantage to RISC/EPIC-based servers. The clustering solution ServiceGuard, for instance is way ahead of what SuSE HA or the Red Hat Cluster Suite can dream to become in 2 years. Though it is possible to reach high levels of availability for the database stack on Linux with Oracle RAC (Real Application Clusters), the management of the solution is quite tedious and its cost prohibitive.

- scalability

Clearly, the biggest deployments of very large databases or ERPs happen on Unix. However, Linux is catching up very fast. Red Hat 6 supports the 120 logical CPUs and 4TB of RAM of the brand new HP DL980 ! Especially when it comes to scalability with regards to virtualization for massive consolidation scenarios, Linux does an excellent job with the compelling argument of the full binary compatibility with the x86_64 architecture.

- ease of maintenance

The number of updates is far less frequent for Unix (at least for HP-UX that I know well) compared with Linux. Moreover, tools such as Dynamic Root Disk, which allows to relocate the operating system from the local disks to the SAN in one command line and one reboot, or to patch the OS on a mirror offline and to reduce the planned downtime to the reboot time, really increase the productivity of the admins. Once again, the tight integration of the operating system to the platform makes this happen.

- partitionning

The partitioning of the RISC-EPIC hardware platform is much more advanced. Because of the capabilities of their platform, x86-based servers rely mainly on hypervisor-based virtualization, whereas Unices offer both hard and soft partitions. Also, although IBM offers Linux on Power servers, most of the virtualization capabilities are only available for AIX, IBM’s Unix. Finally, Linux lacks the pay-per-use and workload management functions of Unices. cgroups implemented in RHEL6 are a beginning, but it still lags behind the HP SRP or Solaris Zones.

A personal note about the “Staffing and skills are not an issue”: my first system administration experience started with Linux and I learned HP-UX afterward. The main principles of *NIX, such as LVM, runlevels, init scripts, the Bash shell, etc. are the same and hence the lack of smart people able to learn a few more commands should never appear. The most important is to stay flexible and ready to learn new technologies !

To conclude, Unix is going to remain in the datacenters for long. Its stability, ease of maintenance and workload management ensure a great continuity in the IT service delivery. Though the number of Linux servers will continue to grow, driven by the increase in performance of the x86_64 platforms (and I am definitely fine with that !), HP-UX, AIX and Solaris (depending on Oracle’s good will) are here to stay…

famillemarcy@ubuntu-home:~$ lspci | grep Net
00:06.0 Network controller: RaLink RT2561/RT61 802.11g PCI

This chipset has a Linux kernel module driver “rt61pci”, but this one did not work. Though iwconfig recognized the chip and ifconfig -a showed the wireless interface wlan0, the following command

famillemarcy@ubuntu-home:~$ lshw -C network

showed the wireless device as disabled.

In order to let it work, I had to add the following parameters in /boot/grub/menu.lst “pnpbios=off  acpi=force irqpoll”. Only after having added it and rebooted it could work.

famillemarcy@ubuntu-home:~$ cat /boot/grub/menu.lst | grep acpi
kernel        /boot/vmlinuz-2.6.32-27-generic root=UUID=853d8b99-8986-4b81-a96d-a635356b0dd1 ro quiet splash pnpbios=off  acpi=force irqpoll

I hope this helps if you have this kind of chip !

What I can say, so far, is that I love this phone ! As already claimed by many bloggers, webOS is brilliant and very intuitive. I can synchronize my emails, calendar and contacts from Exchange. On the top of that, although a proprietary operating system, webOS is based on the Linux kernel, which makes it even more attractive for me and, I think, will make HP an even more important contributor to the Linux ecosystem.

One of the greatest features of this phone is definitely the Touchstone. This station allows to load the phone’s battery through electro-magnetic fields just by putting the smartphone on the station, without having to plug any cable !

I see everyday HP employees from my LinkedIn network activating their LinkedIn application for webOS. I think it is a good thing for the platform to make it attractive for developers. HP’s strategy to make webOS the reference operating system for embedded devices, such as printers, phones, etc. is the right one and I hope to see many more applications for my phone in the future.

This video is not from me but shows how the Palm Pre 2 and webOS look like. Try it and get it !

When I ask my customers how long it takes them to deploy a complete IT service (i.e. architecture design, proof of concepts, physical reception of the servers, provisioning of the resources -such as storage and networking- install the operating system and configure the application) I generally hear answers going from a couple of weeks for small projects to months for more complex projects.

This is the main infrastructure issue the HP CloudSystem Matrix addresses. This Matrix is simply a complete solution made of hardware (c7000 blades enclosure and a wide choice of Intel Itanium-based Integrity or AMD Opteron or Intel Xeon based servers), software (the full stack of the so-called Insight Software Suite) and services (the HP services people come on site, install the hardware and software and customize the solution to adapt it to the customers’ needs and processes).

Instead of deploying applications the old way, HP Bladeystem Matrix really transforms the hardware in commodity hardware that can be used and reused by different services. It creates a shared-service infrastructure that can be divided in different pool of resources which users will pick in order to request a service to be deployed. We can call this a cloud infrastructure in a box, because the users request applications through a self-service portal and these applications are deployed automatically somewhere, on an highly scalable IT infrastructure the business user has no idea about – and that is not his job anyway !

To be concrete: everything starts with the Insight Orchestration Designer. In this web-based application, it is possible to create so-called “cloud maps”, or templates, which will contain all the important information to deploy IT services. Let’s say you want to deploy a SAP 2-tier infrastructure: you will need a database server and some application servers, all booting from the SAN for more flexibility, as well as some network connectivity, i.e. VLANs that will be help deploying the application and the operating system over the network through PXE-boot and also, most importantly, a network linking the application servers to the database server.

How long and in which time frame do you think your company will be able to deploy all these elements manually and individually ? If you think what you have today is not good enough, HP CloudSystem Matrix can help you.

You can create your own cloud map or pick the maps already available on the HP website. The advantage of these pre-prepared cloud maps ? They are designed according to the best practices from HP and the ISVs, such as Oracle, SAP or Red Hat, so you can be sure that you will have enough I/O resources, for instance, to accommodate the deployment of a database server.

On this example, you see the two deployment networks for PXE-boot (RDP for HP Rapid Deployment Pack for Linux on x86-64 and Ignite-UX for HP-UX). The elements, such as server groups or physical storage are picked from the top left corner, drag-and-dropped on the main surface and connected to each other. Creating such a cloud map is really easy, however, this is just a XML file, there is no on-the-fly deployment.

Then, each of the components, such as servers, storage or networking must be configured to enter the minimum requirements needed. At deployment time, the software will pick the elements from the pool of resources which meet at least these requirements. In the case below, we see that the database server will at least have two processor cores and 8GB of RAM.

Note also that you can enter the cost for physical and virtual servers, IP addresses and GB of storage. This helps give an idea of the cost of an IT service to business users. This cost can also be integrated into other applications through a REST API.

We then configure the two network interfaces defined for the HP-UX database server. The primary interface will be the one deploying the operating system (the Ignite-UX network). We then have the choice between a DHCP, static or “Automatic” IP address.

DHCP is the standard dynamic host configuration protocol and “static” will allocate a static IP address to the interface. The trade-off however, is that we won’t be able to re-use the cloud map into multiple services, since the IP address will already be taken.
The field “Automatic” resolves this problem in that the management software will take an IP address from a pre-defined range and will allocate it as a static IP address for the duration of the service. The server, when it restarts, for instance, will keep the same static IP address. When the service is erased, the “Automatic” IP address will be sent back to the pool and can be reused for other services.

Finally, the hostname must contain a hash (#) sign. The reason is that, again, this template can be deployed multiple times and the hostnames need to be different at every deployment. At deployment time, the #-sign will be replaced by a completion string given by the service requester.

We now have to select the operating system that will run on the selected server. In our case, we use HP-UX 11i v3. The x86_64 servers can use Red Hat Enterprise Linux or Windows. Virtual machines running on the top of VMware or Microsoft Hyper-V can also be included in these cloud maps and be provisioned and deployed automatically.

After the main characteristics of the server were defined, we can now make use of the workflows to finish the cloud map. In our case, the workflow “SAPInstall” will be executed at the end of the creation of the service. This means that the SAP instance will be deployed on our newly provisioned infrastructure.This technology is based on HP Operations Orchestration and allows to customize the deployment of the service down to the smallest detail (e.g. change of a kernel parameter,etc.)

At that point in time, it is possible to create your own workflows, such as an integration with HP’s Universal CMDB to update it with the service, or open an new Request for change (RfC) in HP Service Manager or with BMC’s Remedy. The customers can hence integrate the deployment of their services in their already existing ITIL change processes.

Our cloud map is now ready to be deployed. The logo next to “validation status” is now green, which means that we have enough informations to deploy this service. We now save it and tick the box “Published”, so that business users will be able to select this cloud map from their service catalogue and request it to be deployed automatically.

The creation of a cloud map was an easy thing. Remember, though, that HP, in co-operation with ISV’s provide cloud maps ready to be deployed. In my next post, I will show you how to request a service to be deployed and what happens behind the scenes. Stay tuned !

All in all, it was definitely  worth attending this event. Given my role at HP, I regularly participate in similar events and I have found it well executed. Not all sessions were equal both in their content and delivery, but the informations I have gathered in Boston will definitely help me better understanding and managing Red Hat and JBoss on HP servers.

The slides of all presentations are now available. Enjoy !

The second presentation was made by the technical director of the JBoss project at Red Hat about tuning of JBoss.

I then attended a presentation about SElinux that was unfortunately not technical enough and did not meet my expectations.

However, the two last presentations (actually, one presentation filling two time slots) were absolutely fantastic. One kernel developer and a guy from Red Hat’s performance team have made a presentation about tuning Red Hat. It could not go further, as far as the technical level is concerned, and the questions from the audience were also interesting.

Finally, the day ended with a party in the Faneuil hall in Boston and in the surrounding pubs.

The first day started with a good keynote speech from Red Hat’s CEO, Jim Whitehurst, who talked about the shift that software companies will have to make in the future towards more modularity and more openness. He also mentioned how customers could get their money trapped in IT projects for several reasons: complexity, project failures, etc.  It was not a revolutionary speech, but rather a well executed introduction to Red Hat’s vision for the future.

Then came Paul Cormier, Red Hat’s VP for products and technologies who did the dirty job and first explained how Oracle wants to lock customers in their proprietary model with Solaris, Weblogic and the Oracle database.
Wait a second, isn’t that exactly what IBM is trying to do with AIX, Websphere and DB2 ? Of course yes, but IBM being the top sponsor of the event, Mr. Cormier preferred to target Oracle. Fair enough. Then Mr. Cormier tried to show us how Microsoft, VMware and Novell were offering closed proprietary solutions. Even though Red Hat remains a model in terms of a company making money out of open-source software, I do not buy what Mr. Cormier said about Novell and VMware. Novell remains a strong Linux vendor that is committed to open-source software and VMware, even though it is based on a proprietary and costly model, does an interesting job from a standards perspective: they have backed the OVF standard for virtual machines and have acquired SpringSource, a company selling free software.

Finally, a Cisco VP came to tell us how simple the world is with UCS. The more Cisco presentations I am hearing, the more I find their architecture complicated and their rip-and-replace-and-buy-new-switches approach brutal. But I have never seen their systems in action so far.

Following the keynote, I chose the 2 hours lab on Red Hat Enterprise Virtualization (RHEV). Even though the management interface is not as polished as VMware Vcenter (it lacks, for instance, the storage view) and even though storage online migration is not available, it will, in my humble opinion, cover the needs of 85% of the customers for a much lower price.

I then attended a session about achieving best I/O performance, both storage and networking, in a KVM virtualized environment., followed by a great presentation about how to setup a clustered environment with Red Hat 6. Even though RHEL6 is still in beta, the cluster management tool looks really promising.

Finally, I attended two sessions: one about SOA architectures in the cloud and RESTful applications with JBoss SOA platform which were fine.

To conclude the day, Red Hat organized an on-site party with a barbecue. It was interesting to see Jim Whitehurst, Red Hat’s CEO, be there and discuss very simply with the attendees.

All in all, it was a very interesting day and I am looking forward to seeing tomorrow’s presentations!