Review of WA25: Best Practices for deploying the BlackBerry Enterprise Server with Microsoft exchange.
I was recently at WES and got the pleasure to sit in on a session co-hosted by Microsoft. The takeaways from this session were to give best practices for Exchange/BES, dispel some rumors, support concepts behind upgrading to Exchange 2010 and go into detail about the interaction between Microsoft Exchange and the BlackBerry Enterprise server. They opened the session by asking the audience a simple question. Compared to outlook, how does a BlackBerry measure up when it comes to taxing exchange? They had some hands at 3/1, some hands at 5/1 and some at 10/1. At that point, they explained that a BlackBerry taxes the exchange server the exact same way that outlook does. After showing some examples and evidence, it became very apparent this was going to be a very eye opening session.
It seemed very apparent that they were going to push home the fact that the BlackBerry client was having the same impact on MS exchange that an outlook client was having. Now the first thing that concerned me was this was not some misconception; this is something that RIM had said in the past. They then went on to explain that in their past data they had assumed the worst-case scenario. They pointed to the fact that the database IOPS was equivalent, on a per user basis to Microsoft outlook, no 3x, or 5x as prior rumors and thoughts.
At this point, they opened the session up to Robert, who talked about shifts in the industry and some opportunities and challenges that the people at Microsoft and RIM were coming across. Since Exchange 2003, disk capacity (through both cost and size of drives) has grown dramatically. Sequential throughput increasing linearly based on areal density. A point Robert had made was that they did not really expect I/O performance to improve substantially and that at this time, they think that 15k RPM (which in my opinion, is still blazing fast) to be the ceiling.
The industry is quickly shifting in the pro user direction. Users are sending more attachments; users are going to want larger mailboxes. From a display of hands in the room, it seemed like a 1 gig mailbox was the standard, but it seems as if they trend is going to be towards moving mailbox limits upwards to the 10gig size and forward. Employee’s want everything online and searchable. They wanted increased message size, less effort organizing and maintaining your mailbox.
From here, Jeff was introduced to talk about some Exchange concepts. He opened with a quick explanation of ESE which is Extensible Storage Engine, Transaction-based database engine used by Exchange, Active Directory, various other Windows components ( http://msdn.microsoft.com/en-us/library/ms684493%28EXCHG.10%29.aspx) He went into detail about how the data is written in exchange. An interesting point that he made is that transactions are actually written to memory, then written to disk, as obviously it is much quicker to pull from memory then from the disk. Pulling from memory has no tax, versus disk I/O. This typically uses a large amount of RAM. Changes can exist in ESE cache and transaction logs for a period of time before modified pages are flushed to disk (bounded by checkpoint depth) If something is missed in Cache, results in it being read from the db sitting on the disk.
The next slide that Jeff got into was how they minimized Exchange I/O from 2003, to 2007 and even in 2010. The first point was that they had moved into a 64-bit address space so they now had a large database cache. He also made a point to explain that they were moving to larger writes when they did have to write to the database. They are moving towards the idea that they want to have less writes to the database. In the future, they want to have increased checkpoint depth, which means that they do not want to write the same information repeatedly. Some other improvements are
• Increased checkpoint depth
• Repetitive writes of the same page
• DB write smoothing & throttling = reduced transaction latency
• Exchange store database schema
• Lazy view updates = sequential I/O for
• Cache compression = more effective cache utilization
Harping on the point that Microsoft has made significant changes from 2007 to 2010, Jeff pointed out that just because something held true in exchange 2003 or 2007, that it might not hold true in 2010. They gave us a sample DAG deployment, which explained some of the concepts that he had been talking about in previous slides.
