dimanche sept. 22, 2013

T5-2 for high-performance financial trading

On this post, I will focus on the second testimony reported by François Napoleoni. Here,the goal was to select the next best platform to face the growth and upgrade of a financial trading application. The comparaison has been made between :
  1. T5-2, running Oracle VM, Solaris, Sybase and the financial application
  2. x86, running VMWare, Redhat, Sybase and the financial application

The decision criteria being :

  • Simplified architecture
  • Systems performance in real life (what has been tested and measured)
  • Platform stability
  • Single point of support
  • Leverage internal skills
  • Strong security enforced between the different virtual environments
  • ROI of the solution

For those of you understanding french, I will let you listen to this few minutes video below. And for the English readers, go through more details in this post.

lundi sept. 09, 2013

SPARC Showcase in Paris : September 17th

I recently posted about preparing Oracle Open World. A good start, for those of you in Paris, on September 17th, would be to come to the SPARC Showcase, where our customers will develop why and where they are leveraging latest SPARC technologies T4, T5 and Oracle SuperCluster to their IT and Business benefits : 

  • Mr Jean-Marc Jacquot, from Mysis
  • Mr Jean-François Charpentier, Technical Architect from a leader in RH services solutions
  • Mr Sébastien Flourac, Head of Strategy for GRTgaz IT

You can register here.

jeudi juin 13, 2013

3 minutes video of last Month Oracle Solaris Users Group event

A quick report of last month Oracle Solaris Users Group event in Paris... in french...

vendredi mai 17, 2013

Why OS matters: Solaris Users Group testimony

Wednesday evening, a month after the new SPARC servers T5 & M5 launch in Paris, the french Solaris users group, get together to get the latest from Oracle experts on SPARC T5 & M5, Oracle Virtual Network, as well as the new enhancements inside Solaris 11.1 for Oracle Database. They also came to share their projects experiences and lessons learn, leveraging Solaris features : René Garcia Vallina from PSA, did a deep dive on ZFS internal and best practices around SAP deployment and Bruno Philippe explained how he managed to consolidate 100 Solaris servers into 6 thanks to Solaris 11 specific features.

It was very interesting to see all the value that an operating system like Solaris can bring. As of today, operating systems are often deeply hidden in the bottom layers of the IT stack, and we tend to forget that this is a key layer to leverage all the hardware innovations (being new CPUs cores, SSD storage, large memory subsystems,....) and expose them to the applications layers (being Databases, Java application servers,...). Solaris is going even further than most operating systems, around performances (will get back to that point), observability (with DTrace), reliability (predictive self healing,...), and virtualization (Solaris ZFS, Solaris Zones & Solaris Network Virtualization, also known as project "crossbow").

All of those unique features are bringing even more values and benefits for IT management and operations in a time of cost optimization and efficiency. And during this event, this was something that we could get from all the presentations and exchanges.

Solaris and SPARC T5 & M5

As Eric Duminy explained in the introduction of his session on the new SPARC T5 & M5, we are looking into new paradigm of CPU design and associated systems. Following Moor's law, we are using transistors in completely new ways. This is no more a run for frequency, if you want to achieve performance gain, you need more. You need to bring application features directly at CPU and Operating System level. Looking at SPARC T5, we are talking about a 16 cores, 8 threads/core processor, with up to 8x sockets, 4 TB RAM, SPARC T5-8 server in only 8 rack units ! This mean also, 128 cores and 1024 threads, and even more for the M5-32, with up to 192 cores, 1536 threads, 32 TB RAM  ! That's why the operating system is a key piece that needs to be able to handle such systems efficiently : ability to scale to that level, ability to place the process threads and associated memory on the right cores to avoid context switch, ability to manage the memory to feed the cores at the right pace.... This is all what we have done inside Solaris, and even more with Solaris 11.1 to leverage all this new SPARC T5 & M5 servers, and get the results that we announced a month ago at the launch.

 Of course we don't stop there. To get the best out of the infrastructure, we are designing at CPU, system and Solaris level to optimize for the application, starting at the database level.This is what Karim Berrah covered in his session.

Solaris 11.1 unique optimizations for Oracle Database

Karim's explained first the reasoning behind the complete new virtual memory management of Solaris 11.1, something that benefits directly to Oracle Database for the PGA and SGA allocation. You will experience it directly at database startup (twice faster !). The new virtual memory system will also benefit to ALL your applications, just looking for example at the mmap() function which is now x45 faster (this is what is used for all the shared libraries). Beyond performances, optimizations have been made on security, audit, and management. For example, with the up coming new release of Oracle Database, you will be able to dynamically resize your SGA and also get greater visibility for the DBA in datapath performances thanks to a new DTrace table directly available inside the database: a tight integration between Oracle DB and Solaris unique features.

Alain Chereau one of our performance guru from EMEA Oracle Solution Center provided his foresight and expertise. He especially reminded that the performance is achieve when ALL the layers work well together, and that "your OS choice has an impact on the DB and reverse. Something to remember for your critical applications." Alain closed the session with a final advice on best use of SSD for Oracle DB and Solaris ZFS. In short, SSD are align on 4k block. For Oracle DB, starting with, redolog can write in 4k block. This needs to be specify at redolog creation on the record size setting. For Solaris, ZFS knows about SSD and directly adapt. That's the reason why putting ZFS secondary cache on SSD (readzilla) is a very good idea, and a way to avoid bad behavior introduced by new "blind" storage tiering when combined with ZFS. Just put SSD drives for ZFS secondary cache directly inside your T5 or M5 servers and you are done. This is an important topic, as even if a majority of customers are running Oracle Database on ASM on production to get the benefit of grid and Oracle RAC security and scalability, that maybe different for development environments. As a matter of fact, for development systems most customers are leveraging Solaris ZFS and its compression and infinite clone and snapshot functions.

This brings me to René's session on SAP on ZFS...

Lessons learn from deploying SAP on ZFS

Clearly one of the most technical session of this event. Congratulation to René for a very clear explanation on ZFS allocation mechanisms and algorithm policies. I will start by René's conclusion : "Don't follow your ISV (SAP in this case) recommendations blindly". In fact, PSA was experiencing performances degradation and constant I/O activity even with very few transactions on application side. This was due to the fact that SAP recommends to use the SAP Data filesystem at more than 90% ! A very bad idea when you put your data on a Copy-on-Write (COW) filesystem like ZFS... Where I always recommend to keep around 20% of free space to allow for the COW operations to take place ! That's of course the new rule for SAP deployment at PSA.

So if you already have ZFS deployed with this rule in place, you don't have to read further, just keep doing it and move directly to the next topic... otherwise you maybe facing currently some performance problems as well.  To identify which of your ZFS pools are facing this situation, René provided a nice dtrace command that will tell you :

# dtrace -qn 'fbt::zio_gang_tree_issue:entry { @[pid]=count();  }' -c 'sleep 60'

Then to solve the problem, you understand that you need to add free space to enable the COW operation (in one shot). The best way would be to add a vdev (for more details: Oracle Solaris ZFS: A Closer Look at Vdevs and Performance). You could also use a zfs replace with a bigger vdev, but that's not the best option in the long run. If you go through a whole modification cycle of the content of the pool, your zpool will "defragement" by itself. If you want to "defragment" the zfs pool immediatly, if you have a Database, you can do it through "alter table move" operations (special thank to Alain Chereau for the tip). For standard files, you need to copy them and rename them back, or best, do a zfs send | zfs receive to another free zpool and you are done.

From 100 Servers to 6 thanks to Solaris 11

Last but not least, we also had another deep dive session during this event, with live demo ! Thanks to Bruno Philippe, President of the French Solaris Users Group, who shared with us his project of consolidating 100 servers, going from Solaris 8 to Solaris 10 into 6 servers with minimal to no business impact allow ! Bruno achieved his project thanks to Solaris 11 unique new feature : Solaris network virtualization, combine with Solaris Zones P2V and V2V, and SPARC Hardware hypervisor (Oracle VM for SPARC, known also as "LDOM", or Logical Domain).

I invite you to visit Bruno's blog for more details : Link Aggregations and VLAN Configurations for your consolidation (Solaris 11 and Solaris Zone)

Awaiting his next entry explaining the detail of the V2V and P2V operations that he demonstrated to us live on his laptop through a Solaris 11 x86 VBOX image.

I hope to see you on the up coming Solaris and SPARC event to share your feedback and experience with us.

The up coming Paris events will take place on June 4th, for  Datacenter Virtualization, focus on storage and network, and July 4th for a special session on new SPARC servers and their business impact.

mercredi avr. 10, 2013

IT Modernization: SPARC Servers Engineering Vice President in Paris

Avec le renouvèlement complet des serveurs SPARC annoncé il y a 2 semaines, Masood Heydari, vice-président de l'ingénierie SPARC sera à Paris le 18 Avril, afin de partager avec vous, les apports de ces nouveaux serveurs T5 et M5 sur le marché. Après l'intervention de Masood, Didier Vionnet, ACCOR vice-président du back-office, Bruno Philippe, président du groupe français des utilisateurs de Solaris, Renato Vista, CTO CAP Gemini Infrastructure Services, Harry Zarrouk, Directeur des Systèmes d'Oracle pour la France et moi-même, participeront à une table ronde sur les apports de ces innovations pour la modernisation des systèmes d'informations et les nouveaux besoins métiers des entreprises. Je vous invite à vous inscrire à cet évènement afin de venir échanger avec l'ensemble des intervenants.

With the complet renewal of SPARC Servers announced 2 weeks ago, Masood HEYDARI, Senior Vice President of SPARC Servers Engineering will be in Paris on April 18th, to share what the new SPARC Servers T5 and M5 bring on the market. Following Masood key notes, Didier Vionnet, ACCOR Vice-President of Back-office, Bruno Philippe, President of French Solaris Users Group, Renato Vista, CTO CAP Gemini Infrastructure Services, Harry Zarrouk, Director of Oracle Systems for France and myself, will exchange on the benefits those innovations bring to IT Modernization and Business needs.

mardi mars 26, 2013

New SPARC Servers Launched today : Extreme Performances at exceptionally low cost

It will be very difficult to summarize in a short post all the details and already available customers and ISVs results leveraging Oracle investment, design and ability to execute on SPARC servers complete renewal with not one but 2 processors launched : SPARC T5 and M5. It is somehow captured in the title of this entry, in Larry's own words: "extreme performances at exceptionally low cost". To give you a quick idea, we just announced 8 552 523 tpmC with 1x T5-8 (a new 8x sockets T5 mid-range server). Adding on top, "extreme scalability with extreme reliability", as with the M5-32 server, we can scale up to 32x sockets and 32 TB of memory, in a mission-critical system.

New way of designing systems

As what John Fowler was saying : "this starts with design". Here at Oracle, we have a new way of designing. Historically systems were designed by putting servers, storage, network and OS together. At Oracle we add Database, Middleware and Applications in the design. We think what it takes for the coherency protocols, the interfaces, and design around those points... and more.
Today we introduce not one but 2x processors with the whole family of servers associated with them. Thanks to common architecture they are design to work together. All of this of course runs Solaris. You can run Solaris 10, Solaris 11 and virtualize.  No break on binary compatibility.

Direct benefit for your applications... at lowest risk... and lowest cost

This is good for our customers and our ISVs, enabling them to run their applications unchanged on those new platforms with no equivalent performance gain, lowest cost and lowest risks, thanks to the binary compatibility and the new servers design under Oracle era. There was many customers examples on-stage, so I will just pick 2, SAS moving from M9000 to M5-32 with a x15 gain overall, and Sybase moving from M5000 to a T5-2 with a x11 gain overall. Those being in my opinion very important as they are reflecting real applications and customers experiences, many of them being in the financial services, and already having jump on those new systems (thanks to the beta program).

To get a better idea of what the new SPARC T5 and M5 will bring to your applications, being Siebel, E-Business Suite, JD Edwards, Java, or SAP... Have a look here : https://blogs.oracle.com/BestPerf/ on the 17 world records... on performances and price.


Eric Bezille


« avril 2014