[Video] What is the impact of reduced Disk I/O on Exchange 2010 Architecture design?
There is a lot of talk about the performance improvements in Exchange 2010- in particular the reduction in Disk I/O.
Is this real or imagined? What are the tradeoffs Microsoft has had to make in order to make this work?
Join Exchange Expert and Mimecaster Barry Gill talk through some of the issues when looking to design your Exchange 2010 environment to take advantage of the improved performance.
Justin Pirie (JP): So we're back with Barry Gill, welcome to the Mimecast video blog. So, I've been hearing a lot of talk about the performance improvements in Exchange 2010, can you talk me through some of those please?
Barry Gill (BG): Well certainly Justin, I think one of the most significant performance that Exchange 2010 has introduced to us is the 70 to 75 percent input output or I/O performance improvement.
JP: So that's disks isn't it.
BG: That's disk, yeah. Now what Microsoft have done is something they've been trying to get rid of since essentially Exchange 2000, is they've gotten rid of Single Instance Storage, so no longer are read and write operations are being scattered across platters on a disk, they are now, all of these processes are happening sequentially. So sequentially means there's a lot less distance for the heads to travel across the drives which means they're able to get these massive performance improvements.
There are some things that need to be taken into consideration when looking at this. Yes, you can use lower cost, slower speed disks because of this performance improvement but you are also going to experience an amount of bloat on your information stores, Microsoft say that's around 20 percent, which doesn't sound too significant but when you start looking at if you want to do replication through Database Availability Groups, across multiple locations, you're going to have 3 copies of this data, multiply that out by 20 percent and you know every chunk of data is become a lot more.
Plus then if you have multiple sites you're replicating across there's going to be an impact on your Wide Area Network (WAN) links in order to carry that traffic from one place to another.
JP: So I'd heard that it was actually quite a bit greater than that, so 40 to 50 percent greater storage required in some instances, so while its great for future, having cheaper or slower disk for the same performance, do you not think a lot of people have got investments in high performance SAN's where their current Exchange infrastructure is. And what's going to happen- how does that work? Do you extend your current expensive infrastructure, or do you and replace that with cheap infrastructure? And then you need at least, well say 20 to 40 percent more (storage), what do you do?
BG: Well I think every organisation is going to have to handle that depending on what their available resources are. You know the reality is I don't think anyone is going to want to use expensive storage if they have low cost storage available, but if they have to go out and spend money to buy low cost storage and have available high cost storage, they'll probably just use that. The problem comes in and when you start looking to having multiple copies of this information, stored in all these locations. You know, if you're using high cost stuff when it comes time to migrate your data, are you then going to have multiple data migration projects that are going to move data from high speed environment to low speed environment, from server infrastructure one to server infrastructure two? You're creating a level of complexity, and a level of work that is going to have to be undertaken at some point which is obviously going to be daunting for some organisations.
JP: So for me this represents a double gotcha, because not only do you have to have more storage, OK it can be slower, but then you're also shipping more data, because with Transaction Log Shipping you're actually shipping all the data between all the sites, it's actually going to be considerably more than it's ever been before if you want WAN replication.
BG: Absolutely, Absolutely. And the more sites you have the more replication there is going to be. But networks tend to get faster over time, so Microsoft does believe that this is not going to be an issue for many organisations, although I'd like to see them foot the bill for all those IT managers out there who are going to have to buy new switches, considering that every Database Availability Group server, so every mailbox server that participates in a DAG has got to have a minimum of two network interfaces. You know- one for the Transaction Log Shipping and one for mailbox access, your MAPI connections and all that type of stuff.
JP: Cool, thank you Barry.
BG: It's a pleasure.