|
|
| Tuesday, February 09, 2010 |
|
|
|
|
|
|
|
 |
|
 |
| |
Written By: Bernd Harzog
On May 28th 2008, VMware announced that it was acquiring B-hive. My initial take on this acquisition is covered on my posting for that day. The short answer is that I think this a great move for VMware. Enterprise customers need to be able to know how well their virtualized applications are performing for two reasons. One is that they need to be able to provide assurances and service level guarantees to line of business executives in order for the business to be comfortable with the virtualization process. The second is that this comfort level is essential in order for enterprises to extend virtualization to more business critical applications.
With this acquisition, I think the following dynamics have been created:
- VMware (the market leader) in virtualization platforms has acknowledged that when applications move from dedicated servers often owned and funded by the business, to shared servers owned and operated by IT, that IT needs to take more responsibility for applications performance than IT previously took.
- VMware has acknowledged that Applications Performance Management for virtualized applications is a different problem than for physically hosted applications (it is, see my blogs on the subject), and that virtualization aware APM must be used in virtualized environments to both ensure good performance, and drive further virtualization more quickly.
- VMware is sending a message to its customers that application response time is THE metric that should be used to both manage capacity, and manage applications performance.
- APM for virtualized applications has now moved from a "maybe" to a "must have".
- Changing how APM for virtualized applications systems is perceived raises the bar for vendors of competing virtualization platforms. Citrix and Microsoft must now either form partnerships, or ultimately do their own acquisitions in this space in order to keep pace with VMware.
- A mad scramble will now ensue among the remaining APM vendors for favorable positioning in the virtualization space. The rest of this blog is about how the APM industry is impacted by this set of dynamics, and who the likely winners and losers will be.
B-hive - A Quick Look at the Product
The basic idea of an APM or EUEM solution is to measure either applications performance (how fast is the application processing transactions or units of work as they arrive from users or other applications), or the end user experience, which measures how well the application is working in the eyes of its end users. B-hive is not a true end user experience solution since in order to be one of these you have to be able to measure response time from the perspective of actual end users, and B-hive does not do this. B-hive measures response time from the perspective of the presentation tier (the web server if it is a web based system) of an application system, which is exactly the right approach if you want to be a leading edge APM solution (older APM solutions tried to infer applications performance by looking at the resources used by an application, and this approach simply does not work for virtualized applications systems). So, here is a quick overview of what B-hive does:
- B-hive attaches to a mirror (spanned) port on the switch that supports the servers closest to the users (most often the switch that supports the web servers). This allows B-hive to see all of the transactions that flow back and forth between users and the applications system. In a VMware system, B-hive is implemented as a virtual appliance that attaches to the mirror port on the virtual switch inside of the VMware host.
- B-hive's concept of response time is the time between the arrival of a request from a user, and the reply to that request on the part of the application system. By default these transactions are "atomic" or highly granular and not something that that map to what a user would view as a transaction of interest. But they are representative of the overall response time of the system, and since these atomic transactions can be supported for almost every application with no configuration, they represent the right level of detail for the IT staff supporting the virtual infrastructure.
- B-hive does a very good job with HTTP based (web based) applications, and is very adept at discovering the structure of the application through an analysis of the HTTP data flow of the application.
- B-hive also understands certain back end protocols like SQL Server and Oracle. So for two-tier client server applications or any application that talks to a database, the request/response time to and from the database server is used as a proxy for response time.
- B-hive also has built application level protocol decodes for applications that are not HTTP based. For example, B-hive contains a decode for the RPC protocol that is used between Outlook clients and Exchange Servers.
- While B-hive does a great job with web based and database based applications, it does not support all TCP/IP based applications at the applications layer like it does with HTTP based applications, and the specific applications for which protocol decodes have been built. For applications that do not fall into the web or specifically supported category, the best fallback is to look at how fast the database is responding to requests.
- With additional effort, atomic transactions can be combined into true compound transactions or transactions of interest to applications owners and the business. In fact B-hive is sold in two flavors, one that supports just atomic transactions for the IT staff, and another one that includes the more comprehensive view of higher level transactions which are of interest to applications owners and the business analysts.
- B-hive will try to get the user ID of the application out of the TCP/IP data stream. For web based applications this is easy. For some applications this is not possible, and the notion of who the user is defaults to the source IP address which is in many cases not unique to a user (due to NAT). This is one area where vmSight has an advantage due to its patented Connector ID technology.
- B-hive claims to be able to automatically drive actions in the virtualized environment (for example provision a new server or move a VM) based upon its response time measurements. This takes DRS to an entirely different level, since making these kinds of decisions based upon applications response time makes much more sense than does making them based upon CPU or memory usage.
Post B-hive - APM Winners and Losers
Prior to this acquisition, a large number of monitoring and management vendors had either positioned their entire companies as virtualization management plays, or positioned some of their products as being relevant to virtualization in some way. The chart below, maps groups of vendors, and some specific vendors on their value in the virtualization market based upon two factors. One is the breadth of their applications support (from a few applications to all applications). The second is the degree to which the product can actually measure true user experience within an application. This is a tradeoff that needs to be made, as there is currently no product that can measure transaction response times within an application without per transaction configuration, which means you cannot have per transaction response time if you want a product to support all of your applications out of the box (with no configuration required). The reason configuration is an issue is that most enterprises have over 1000 business critical applications, and no enterprise has the staff or the budget to instrument N transactions in 1000 or more applications manually.
The Losers
The losers in this equation are vendors that primarily rely upon a resource usage approach to APM (measuring how much CPU an application is using), and vendors that primarily collect data that VMware Virtual Center already collects and just regurgitate it. APM via resource management (see my Solutions Guide White Paper on why) is a discredited approach within virtualized systems since time based data collected within the VM gets corrupted by the virtualization process. Therefore the losers are:
- Management Frameworks from IBM, CA, HP, BMC, and CompuWare. These products are used to measure the availability of the entire IT infrastructure. They were widely perceived as being cumbersome and expensive before virtualization. The dynamic nature of these environments combined with the time shifting for resource based metrics, made these products marginally useful for the management of virtualized systems. Now the VMware has acquired a vendor with an application response time approach these products look even less attractive since they combine a discredited approach with their existing problems of being cumbersome and expensive.
- Resource Based APM Products (BMC, NetIQ, eG Innovations, Hyperic, and everyone who collects resource data from Windows Servers "agentlessly" via WMI). These products take the wrong approach to APM (measuring resource usage on a per application basis). How that VMware has acquired a vendor that does it the right way, this approach has even less traction in the market than it did before.
- Some VMware Focused Startups. Many management startups have tried to position themselves as "the virtualization management vendor", and by extension position themselves as the management vendor for VMware. Many of these startups do nothing more than take existing data out of Virtual Center or the VMware SDK, and repackage it. Examples include VisionCore, VKernel, and many others. Since these vendors take an approach that is superceeded by B-hive and since they have no plausible story to add value to VMware, they are losers in this equation.
The Winners
There are quite a few different winners in this equation, and each one is a winner for a different reason. So, they are listed below individually:
- Akorri Networks. Akorri collects data from the SAN and the virtualized servers, and calculates performance and capacity utilization from the perspective of the infrastructure. While the chart above says that end user response time is more valuable than infrastructure response time, this is true only if you do not have either. Once VMware customers deploy B-hive, the very next question they will ask is "Thanks for telling me it is slow, now, why is it slow?" Akorri can answer this question today because it collects the data and then has the built in analytics to figure out where in the data the problem resides. Once VMware makes the B-hive data available for third party vendors to consume, Akorri will directly benefit from this by being able to pull application response time data from B-hive into its model, and use the model to provide an end-to-end root cause mechanism for APM. In other words, Akorri is perfectly positioned to provide the customer with the missing piece that completes the value of the B-hive deployment. Akorri will also benefit by supporting VMware, Citrix and Microsoft virtualization platforms, giving customers who end up with more than one virtualization platform a cross-platform tool.
- vmSight. vmSight is very similar to B-hive, but there are some important differences. vmSight's notion of "Application Response Time" does not go up into the applicationi as much as does B-hive, but it does int turn support all applications that use TCP/IP. So, the supported set of applications for vmSight is broader than with B-hive. vmSight (like Akorri) will end up supporting all three major virtualization platforms (VMware, Citrix and Microsoft). This again is critical for customers who are or may run more than one platform. Finally, vmSight can tell you exactly who the user is (their actual Windows logon ID). This is done via their patented "Connector ID" technology, and it will be critical to solving the APM problem for VDI deployments. As VDI gains traction, vmSight will win as customers discover that B-hive is not adequate for their VDI implementations. Furthermore, as Citrix and Microsoft gain traction in the virtualization space, and vmSight adds support for these platforms, vmSight will be "the B-hive" that works across VMware, Citrix and Microsoft.
- HTTP Appliance Vendors (CA/Wily, HP, CompuWare, Tealeaf, Coradiant). These products measure the response times of HTTP transactions for web based applications. They need to be configured for each application and each transaction, but for applications that warrant this level of instrumentation, these products were valuable before B-hive and continue to be so. They collect valuable response time data via a method that still works when you virtualize the application, and the data they collect is not collected by B-hive.
- Knoa, Aternity, and any other vendor of application/transaction specific response time monitoring. These vendors put agents on the desktops of the users running certain business critical applications and measure the actual end user response time for specific transactions within the actual application run by the user on the desktop. Since B-hive does not collect transaction specific data, nor does it collect it from the perspective of the actual end users, these products provide both a set of data and a perspective on performance of the application than no one else can provide. If you are virtualizing the back end servers for a business critical application like SAP or Siebel, and you want to know how it really is performing from the perspective of every user of the application then this is what Knoa can provide you. Aternity does not support SAP, but they do support a variety of other important business critical applications.
- IT Correlation Vendors (Netuitive and Integrien). These vendors specialize in taking data collected by other vendors' products, and finding abnormal patterns in the data via self learning correlation or other statistical approaches. Since B-hive will provide a picture of true application response time, and since the B-hive data will at some point in time become available to these third party vendors, Netuitive and Integrien will be able to dynamically correlate all of the infrastructure data to find probable causes in the degradation of application response time. It will be a race between Akorri who already directly collects infrastructure response time data from the SAN and VMware and runs it through their own model, and the IT Correlation vendors who need to access both the infrastructure data and the B-hive data to provide the true end-to-end root cause picture that enterprises need for their virtualized systems.
The End Game
I think that the two vendors that are best positioned to prosper in this new environment are Akorri and vmSight. Akorri, because they have 2/3rd's of the data that they need already (adding the data from B-hive completes the picture), and because they have substantial built in analytics to process this data to provide that end-to-end root cause picture. vmSight will prosper because they are the only vendor of application response time data for all virtualized applications, and because their approach works for both virtualized servers and desktops (B-hive's approach only works for virtualized servers). Look for vmSight to work with VMware on VDI, and with Citrix and Microsoft on both virtualized servers and desktops. Finally, Citrix and Microsoft have to both react to VMware's expansion of the virtualization suite. They will initially do this by forging close partnerships with some of the winners in this article. As Microsoft officially enters the market later this year, and starts to push down price points, the virtualization platform will become commoditized, and all three vendors will have to focus on the management tools that support the platform to drive license revenue. This will cause further acquisitions to occur in this space.
Article Tags
Trackback this Article
http://www.dabcc.com/trackback.aspx?nCdType=1&nCdContent=7929
Latest Articles
|
|
|
|
|
|
|
|
|
|
|
Featured Partners
Partners
Become a DABCC Partner |
|
|
|
|
|
|