Schedule:
For details and registration see Solaris@STEP
We have integrated ZFS into Solaris and therefore through OpenSolaris and Solaris Express (for now the Community Releases starting with build 27) you now may read the documentation and even the source code (well, the ultimate documentation), and you man not just read about it, you also can play with it. ZFS is not just a new filesystem - it takes a new approach to filesystems as it dissolves the the boundary between volume management and file systems. It moves from the traditional abstraction of a filesystem as a fixed structure within a virtual disk towards filesystems which offer just a namespace and dynamically allocate space from a storage pool. Jeff Bonwick, the chief architect for ZFS, had previously designed the kernel memory allocator for Solaris. Sometimes it is important to make a step back and rethink the whole problem not just optimize bits and pieces. E.g. in ZFS we did not optimize file system logging, but we got rid of the need for logging at all by ensuring that the on-disk state is always consitent by never rewriting data but using the copy-on-write paradigm and atomic state transitions instead. ZFS may be considered the last word in filesystems which being a 128-bit filesystem has enough capacity (more than a Zetabyte) to boil the ocean. It's not yet the final product version, e.g. we are still tweaking the performance - it is already great in many cases but will become even better until it will show up in the Solaris 10 product during the first half of 2006.
We have also launched our 2nd generation throughput computing processor codenamed Niagara and now officially termed UltraSPARC T1. This processor can process the highly parallel workloads found especially in the web-tier of datacenters, but also in other tiers, much more efficiently on it's up to eight simplified and less complex cores than traditionally designed processors. The internal utilization of a processor today is often only about 25% as it is stalled waiting for data from memory, the T1 typically has a much higher utilization as it keeps the processing units busy with its four threads per core - and Solaris as an operating system is able to efficiently handle these many threads and can schedule them to the 32 CPUs an UltraSPARC T1 provides - 8 cores each with 4 threads, and each of these shows up as a CPU in Solaris - obviously a real system-on-a-chip (based on its processing power it is an E10k on a chip!). And we now also have launched the systems with these CPUs (actually we have already been shipping them for a few weeks now): the Sun Fire T1000 and T2000 CoolThreads Servers. Today the cost of power consumption of a server calculated over three years is in the same order of magnitude like the cost of aquisition, it still is a factor of 3-4 smaller, but it is in the same order, and while the cost of hardware is declining energy costs are on the rise. And it's not just about cost, it is about eco-responsibility in general. In the launch this week we used a demo called Sim Datacenter to clearly show the value of T1 based systems over our competitors. It is a very simple, easy to use tool to simulate a datacenter and see how the variables of cooling, power, space and performance are affected. Customers liked it so much, they asked us to make it available, so here it is. Interested in more technical details ? Richard McDougall has collected a number of relevant blogs.
By the way: we just made a major step forward in our line of traditional UltraSPARC processsors with the introduction of UltraSPARC IV+. This is much larger step from the US-IV than the small "+" seems to imply. The naming only highlights its compatibility: it fits into the same systems as the US-IV. We moved from a 130nm to a 90nm design, we have a new core which introduces a third level cache, and we have slightly increased the frequency to 1.5GHz. All of this has increased this processor's single core performance by a factor of 1.5-2. We just announced a bunch of performance records archieved with this CPU.
I already had blogged about current shifts in software licensing. All the innovation in Solaris 10 convinced Oracle to select the Solaris 10 Operating System as its preferred open source 64-bit development and deployment environment. They also have recognized the need for change in licensing. For licensing on the T1000 and T1000 servers there is a factor of .25 instead .75 applied per core.
But software business modells are about to change even more driven by the move to open source and free software. Sun is moving ahead of the pack by making the Java Enterprise System, developer tools and N1 software available at no cost. This especially lowers the barrier of entry for developers. Together with Solaris this environment forms the Solaris Enterprise System. And by adding the Postgres database to Solaris 10 it even offers a free relational DBMS. But there is still a business model behind all of this. You man choose to take the free option, then you are on your own or may hope on free community support, but if you require warranty, and need to rely on support then we offer services for a fee. Jonanthan Schwartz explains this very nicely (do you want his home phone number?).
And Open Source is not just about software. With OpenSPARC Sun open sources the implementation of the UltraSPARC T1. While open sourcing our innovations we continue to invest in research and development and thereby drive technology. In an article in IEEE Micro earlier this year Marc Tremblay etc al. describe some details of what we are doing to further improve the next generations of throughput computing chips e.g. by bridging the widening gap between processor power and memory latency through scout threads that bring in data from memory into the cache hierarchy more effectively than todays approaches.
Innovation still matters these days.
I am giving a presentation on Solaris 10 for developers which will be posted by marketing soon. For immediate access here a list of references from my presentation [Update: Slides available now, the list below includes some additions] :
DTrace
Security
Performance
Availability
Update: More Useful URLs
Overview of new Solaris 10 features with many references.
A discussion with a customer reminded me of a
discussion thread in the OpenSolaris DTrace forum on
"a bit of IBM anti-DTrace propaganda"
which especially points out DTrace' superior security model compared to the approach taken in SystemTap -
which is still work in progress for Linux.
The
When I read this headline I wondered why Solaris should be more expensive than Linux, given that Solaris service and support is priced lower than either Windows or Linux - a fact Tom Sullivan found with surprise even in this report. The reason is pretty simple: The report included hardware cost and based their calculation on list prices. For Solaris they only considered the SPARC platform, though it would have been easy to compare the various operating systems on exactly the same x86-hardware, eliminating assumptions made regarding actual pricing. This would have made Solaris really compelling. In addition keep in mind that there is no price for Linux per se, there can be a significant difference wether you pick a SuSE or RedHat distribution. When comparing subscription prices you really need to make sure that equal service levels are compared, as e.g. Novell/SuSE charge extra for phone support.
Another aspect of this study may also be considered a systemic flaw: the focus just on the application server environment. Typically it is part of a larger environment. It makes a difference in TCO wether you run such a multi-tier environment homogenously or heterogenously - backend systems are the domain of RISC-based servers and a homogenous Sun SPARC solution may have a lower TCO than a heterogenous solution combining AIX/POWER and RedHat/x86. Related to this is the cost for administration staff. While such TCO studies often correlate it with the operating system actually it may be more correlated to the complexity of an environment.
On the hardware side there may be major differences among different x86-Plattforms. Power and cooling have become major cost factors these days. Marc Hamilton did an interesting analysis for a large compute cluster comparing Intel- and AMD-based servers.
Over the next months Sun will release products which are going to disrupt existing TCO studies.
Customers also do expect that licensing schemes recognize virtualization technologies which allow to create multiple virtual servers or execution environments within one physical server. They only want to pay for the part of a server which actually is being used. Solaris 10 introduced containers which may be bound to resource pools with only a subset of the CPUs in a system. Oracle recently accepted containers as a mechanism for hard partitioning systems such that Oracle only needs to be licensed for the number of CPUs in the resouce pool.
Virtualization on the datacenter level - like with Sun's N1 Grid requires even more flexible licensing schemes as customers want to move around worklods flexibly within their whole data center - using small or large systems as appropriate for the current workload. Capacity based licensing schemes, especially schemes which differentiate between system categories like workgroup or enterprise servers are not apropriate for such dynamic environments. Licensing schemes based on the number of employees as used for Sun's Java Enterprise System or other higher level metrices are much better suited.
Open Source drives a new business model for software in general: subscriptions. With Solaris 10 Sun moved to such a subscription model for the Solaris OS.
Update: The Economist has a good article on how new computer chips are challenging the traditional structure of the software industry and also Jonathan Schwartz just expressed his thoughts on this topic in his blog
Q: I was told that Sun was going to set up a blog for you.
A: Yeah, (after) the first one I wrote, they said, "Nope, you're not blogging." It would have easily been the most read blog on the planet.
Q: Was it the lawyers who stopped it?
A: No. The way that I blog is not in the classic blogging sense. But I have like 40,000-plus customers and partners who subscribe to my boardroom minutes, which I send out every couple or three weeks or whatever, and it's kind of a subscription as opposed to a click-on kind of (blog). I just do it in a different way.
In May USA Today had an article on corporate blogging in the industry in general.
Yesterday we completed the aquisition of Tarantella. Tarantella's technology and products offer secure access to server-based applications running on Linux, UNIX, Microsoft Windows, and even Mainframes from anywhere on virtually any Java enabled device as a service. This allows for the consolidation of critical applications and data onto centrally managed application servers and off of individual desktop and laptop computers, improving resiliency and increasing flexibility. The Tarantella website lists a range of solutions which have been created based on the products. Looking forward this technology may be used in a utility computing model for the desktop. Service providers, OEMs, device manufacturers should think about such new business opportunities.
Two other aquisitions are still pending: StorageTek - Sun's largest aquisition ever, announced June 2 (see Jonathans's Blog for more background) and SeeBeyond announced June 28.
Another technology aquisition in the data management space just has been completed on June 13th. Sun aquired all intellectual property rights of Procom Technology, Inc. relating to its Network Attached Storage (NAS) offerings. Through a previous software licensing agreement with Sun, Procom's technology is already embedded and shipping within the Sun StorEdge 5000 family of NAS Appliances.
SeeBeyond has already been a partner to Sun in Sun's approach to service-oriented architectures (SOA) - Pragmatic SOA - which emphasizes incremental projects, quick returns on investment, and building out a SOA infrastructure in phases. At the JavaOne Developer Conference Sun announced additional new products and support for new web services specifications for SOA infrastructures. These include the Java Business Integration (JBI) specification (JSR 208) and the first fully open sourced Java Enterprise Service Bus (Java ESB) built on this specification.
Mentioning open source: We are also moving forward with open sourcing key Sun software assets: The Java System Application Server (Project GlassFish) and our web single sign-on solution. Like OpenSolaris both are licenced under the CCDL - and again there are some spreading FUD about Sun's open source licensing.
And of course the JavaONE Developer Conference 2005 June 27-30 celebrated Java's 10th birthday. Ed Burns has put together an extensive conference summary in his blog.
Using Solaris on their laptop is important to many - including myself. An article on opensolaris.org has more on this topic.
On the software side IBM announced at the JavaONE Developer Conference, that they will broaden support of its DB2, Rational, Tivoli and WebSphere software to include the Solaris 10 OS on x64 AMD Opteron-based platforms. Reuters is another large ISV porting to Solaris on x64. If YOU need a software from an ISV who does not yet support Solaris on x86 it will help to state the demand by asking this vendor when they are going to provide their software on Solaris on x86. On the other hand an alternate solution may already be available or in the works. While you need your ISV for closed source applications, most open source applications are immediately available for Solaris 10 on x86. The Burton Group recommends in a presentation Solaris 10 as a compelling alternative to Linux especially for OpenSource workloads.
Many customers already enjoy the superior performance and scalability of Solaris 10 on x64.
To fully exploit the capabilities of DTrace some understanding of Solaris internals is useful. Jim Mauro and Richard McDougall have put a current (March 2005) 536 slides presentation on Solaris internals and performance on their website www.solarisinternals.com which complements their book. While the book is a little dated (reflects Solaris 7) it's still a good basis and you may use it together with the slides mentioned above while they are working on the second edition. In addition their blogs offer a wealth of articles on how to use DTrace to understand what's happening in the kernel.
Alan Hargreaves has put together a nice article on how to use Statically Defined Tracing (SDT) Probes to instrument your own code such that you can use the DTrace framework for monitoring and analysis. SDT is documented in the DTrace manual, but it's a rather thick manual and it's likely that many software developers will miss this promising use case: a powerful instrumentation for your own code tightly integrated with the instrumentation in the OS - something performance analysts will highly appreciate - without generating overhead if not active.
If you are more interested in the latest binaries than in the sources check out Solaris Express. Here you may find what's new in the latest version of Solaris Express - e.g. support for iSCSI - compared to Solaris 10 03/05 general availability version.
P.S.: ... and you may show your enthusiasm for OpenSolaris using a fan button
