Sunday Apr 19, 2009

SPECjAppServer2004 2925.18 JOPS, Glassfish and MySQL raise the bar


SPECjAppServer2004 2925.18 Sun continues performance and price/performance leadership

Good News :

Sun has just published our latest SPECjAppServer2004 results of 2925.18 SPECjAppServer2004 JOPS@Standard using all open source stack i.e. Glassfish , MySQL and OpenSolaris on the new Sun / Intel Nehalem based servers (using ZFS).

Highlights:

  • Nehalem/Glassfish/MySQL combination brings substantial price savings to users of typical web based applications and all with commercial grade support

  • The tested configuration consists of 1 x Sun/Intel Nehalem based X2270 server for the Glassfish application server and 1 x SunFire X4170 Intel Nehalem based server for the MySQL 5.1 database. For more details see the result at http://www.spec.org/jAppServer2004 or look at http://blogs.sun.com/kevink for a good overview.

  • MySQL efficient logging system requires only 12 spindles to achieve the I/O throughput to support this load, yet most of the proprietary software results published require at least double this !

  • First industry standard benchmark featuring the database running on a ZFS files system (which made running multiple tests very efficient , this is a big deal so I will give some more details on this shortly)

  • Total software and hardware purchase price $US 38,880 based on the “Bill of Materials” in the benchmark report and online pricing from sun.com.

  • Significant performance gains i.e. the previous best all open source result was Sun's 1197 JOPS result (see http://blogs.sun.com/tomdaly/entry/sun_compelling_price_performance for details ) but this new result is on Sun/Intel Nehalem servers is more than double the previous best.

  • Despite achieving more than double the score of the previous best all open source result this configuration still yields a price/performance gain over the previous result I.e $US 38,880 / 2925.18 = $US 13.29 / JOPS@Standard and the previous best was $13.46/JOPS@Standard (see http://blogs.sun.com/JenniferGlore/ for pricing details ).

  • To give some idea of the actual performance , this configuration of 2 x 8 Core servers and open source , supports a virtual concurrent user load of more than 22,750 web application virtual users and uses > 800 connections to the MySQL 5.1.30 via the Glassfish connection pools!

The Test rig


Conclusions:

  • If you are a software developer of web based applications , perhaps an ISV, or perhaps a contributor to an open source project then you really ought to consider the price/perform and enterprise level support advantages of the Sun open source stack and start considering certifying or deploying your application to this platform.

  • If you are an end user customer of web based applications such as e-commerce or similar applications you should perhaps ask your software supplier when they can start moving their applications to the Sun open source stack, so as to start saving you and them money (especially in these tight times)

  • If your hardware supplier is not helping to drive your costs down like Sun is, then perhaps start asking them why !

Final thought :

These results using MySQL 5.1.30 don't yet include the performance improvements from the marvellous work that the combined Sun/MySQL/Community performance team have been working on so watch this space, the good news is not finished yet.

Pricing is based on Sun Glassfish Enterprise Server 2925.18 SPECjAppServer2004 JOPS@Standard result Bill of Materials at http://www.spec.org/jAppServer2004                                                                                                                                                                                                       
Required disclosure : SPEC and SPECjAppServer are registered trademarks of Standard Performance Evaluation Corporation.
Sun GlassFish Enterprise Server v2.1 on Sun Fire X2270 with MySQL 5.1 on OpenSolaris 2008.11                                                       Sun result 1  x X2270  8 x cores (2 Chips) used for application server and 1 x X4170 Database server 8 x Cores (2 Chips)   

Thursday Nov 06, 2008

Sun Demonstrates Compelling Price Performance

Sun demonstrates compelling price/performance advantage of enterprise supported Open Source.

Using Glassfish V2 U2 / MySQL 5.0/ OpenSolaris on commodity Sun Intel hardware (i.e. SunFire X4150 Intel based servers)

Sun has achieved a result of 1197.50 SPECjAppServer2004JOPS@Standard , you can see a lot more detail on this result on the Sun benchmarks page and also a nice write up by the bmseer and also checkout the Sun ISV blog here

The Summary :







Why this is interesting/important to organisations running web applications.

  1. Demonstrates the huge cost savings available from using Open Source software for the entire application stack !

    The table and graphs above highlight an important fact. The Sun solution for hardware and software costs less than one CPU's worth of database license as used on the Dell or HP results and this isn't true just of the results above but in fact true of most of the SPECjAppServer2004 submissions that use proprietary (application server or database) software. You can check out the details of the pricing here

  2. Demonstrates Open Source solution competing directly against proprietary solutions

    One very useful and important feature of the SPECjAppServer2004 benchmark is that it is written in java and this makes it possible to run the same code unchanged on a variety of different operating systems and hardware platforms, in fact there is a benchmark rule that prohibits any code changes being made to the SPECjAppServer2004 java code. This means that no matter what platform , j2ee application server, operating system or database the application is run on the application itself is unchanged. This characteristic makes SPECjAppServer2004 and ideal workload for making comparisons between  different environments and Sun is utilising this fact to provide a comparison between the proprietary software based Dell and HP results with the an all open source based result. In addition to running the same java code and hence accessing the same database data, the open source results must fulfill the benchmark requirements for enterprise products such as , correctness, support, durability ACID compliance etc.

    Of course there are limitations with all benchmarks and comparisons must be made carefully, however this result clearly demonstrates open source doing the same job as proprietary but at a fraction of the cost. This coupled with the continuing performance improvements I mention in 3. below seem to be adding weight to Allan Packers thoughts on the topic of the future of proprietary and open source databases.

  3. Demonstrates continuing and outstanding performance improvements in the Open Source stack and in the MySQL DB in particular

    In fact it turns out that the Sun result of 1197.10 JOPS demonstrates the best published “commodity hardware” result for per core database performance. The Sun benchmark uses just 4 Intel cores on one chip in the X4150 database server and this means that the combination of OpenSolaris/MySQL 5 and X4150 delivers per core JOPS of 299.28 JOPS/database core. The only results that demonstrate superior per core performance for the database are some IBM power architecture chips (which are definitely not commodity priced).  If you want to verify this claim please download the spreadsheet of results available via the search feature at http://www.spec.org/jAppServer2004/results/

    Also it turns out that the MySQL database requires far less disk resources that competing products, this is demonstrated by the fact that the internal disks of the X4150 were used to support the database throughput and only 6 of them in a raid 0+1 partition. If you look through the SPECjAppServer2004 results you will see most vendors need to use expensive disk arrays to achieve the required disk performance (and most likely expensive extra database options).

  4. Just the start of performance improvements for MySQL , Glassfish and Open Solaris

    This result is based on the first release/update of OpenSolaris and uses MySQL version 5.0. and the performance of both of these platforms is going to improve in the near future for instance with  MySQL 5.0 we simply have not yet seen  the results of the impressive gains the Sun Performance Technology team in conjunction with the open source community and open source focussed companies such as Google are bringing to MySQL.  We will see these performance gains start to turn up with MySQL 5.1 and beyond (see Allan Packers article on this).  Also stayed tuned as the performance of Glassfish is going to continue to impress especially on commodity 4 and 8 core servers used in conjunction with MySQL. In fact I can't wait to start testing GF V3.

  5. Because this benchmark result can be a pointer to how your organisation can start to save “serious” money by using enterprise supported Open Source software

    Basically you might want to start investigating what web applications in your organisation can run on an enterprise supported open source stack such as the glassfish/MySQL/OpenSolaris stack used for this benchmark result.
    There are some steps you can start talking now and I am going to detail a strategy for doing just this in an upcoming blog.


Required disclosure : SPEC and SPECjAppServer are registered trademarks of Standard Performance Evaluation Corporation. Results from www.spec.org as of 11/05/2008. 2xSun Fire X4150 (8 cores, 2chips) and 1xSun Fire X4150 (4 cores, 1 chip) 1197.10 SPECjAppServer2004 JOPS@Standard; Best result with 8 cores in application-tier of the benchmark: 1xHP BL460c (8 cores,2chips) and 1xHP BL480c (8 cores,2 chips) 2056.27 SPECjAppServer2004 JOPS@Standard; Best result with 2 systems in application-tier of the benchmark: 2xDell PowerEdge 2950 (8 cores, 2 chips) and 1xDell PowerEdge R900 (4 chips, 24 cores) 4,794.33 SPECjAppServer2004 JOPS@Standard.



Saturday Aug 23, 2008

VirtualBox OpenSolaris and Ubuntu Guests on OSX networking working

Finally have networking going on OSX

I have been having trouble getting guest OS networking going with VirtualBox 1.6.4 using a variety of  guest OS's on my mac book pro. I have been trying Ubuntu (Gutsy), OpenSolaris and Sugar OS (from OLPC,the one laptop per child project) as VirtualBox guests  but to date had not been able to get networking functioning properly. I finally have both Ubuntu and OpenSolaris networking and full screen modes working now quite nicely.

The things that really helped me  with VirtualBox 1.6.4 were Dana's blog entry on the subject  and the following points in particular : - 

  • The comment not to rely on ping/ICMP as the VirtualBox documentation apparently makes it clear that this is unreliable.
  • The reader comment at the bottom of this blog saying that the DNS server installed by NWAM of 10.2.3 doesn't work. 

So on both OpenSolaris and Ubuntu I manually edited the /etc/resolv.conf and inserted the value for nameserver that I determined from looking at cat /etc/resolv.conf on OS X. I then bring up the browser (not ping) and I can indeed access the internet and browse the web. I really need to get the Sugar/XO laptop emulator working in VirtualBox so this is my next target.

 Oh the one thing folks might also find useful is that to operate OpenSolaris in fullscreen mode on OSX simply mount the VirtualBox guest additions (part of the download package I think) and do perform a packageadd. Once you have done this logout/log back in again and you can use apple-f to go to full screen and apple-f to get back out.



About

tomdaly

Search

Archives
« July 2014
SunMonTueWedThuFriSat
  
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
  
       
Today