DAS vs. SANs for Exchange

For a while now, Microsoft and its larger consulting partners -- such as HP -- have been telling customers that direct-attached storage (DAS) makes a lot more sense than shared disk arrays for their Exchange mailbox stores. This has been greeted with a certain amount of incredulity in some quarters. Let's scratch the surface ...

As with most email-centric server applications, Exchange has always had difficulty scaling, due to disk performance bottlenecks. As the number of active users grows, so does the number of simultaneous read and write requests to the disk storage. Even the best disks need several milliseconds to service an average request -- that's simple physics -- so the classic solution to high user loads was to spread the load over more and more disks, using techniques such as striping. This is what disk arrays are good at -- fast, multiuser access, while keeping the data safe from inevitable hardware failures.

DAS, on the other hand, is much less expensive, but not as scalable. The classic assumption was that DAS was fine for small, brand office servers, but anything more taxing needed an expensive, high-performance, external array.

With Exchange 2007, however, much more of the disk usage can be serviced by a large in-memory cache. In previous versions, the maximum amount of physical memory that a server could have was less than 4GB -- this limit has been raised substantially in Exchange 2007. According to tests run by HP, moving from 4GB to 22GB of memory can reduce physical disk accesses by 70% -- although you should note that this is a best-case figure.

The combination of this additional caching and the improved performance of disk interface technologies such as serial-attached SCSI (SAS) has substantially changed the economics of storage.

... Richi Jennings

One Comment

  1. Paul
    Posted August 21, 2008 at 4:39 AM | Permalink

    Exchange is not my bag, so maybe I have misunderstood, but you seem to be equating SANs and arrays. There’s nothing to stop you running an external array as DAS, but you appear to imply they are necessarily mutually exclusive: “The classic assumption was that DAS was fine for small, brand office servers, but anything more taxing needed an external array”. Am I missing something? SANs only become interesting (so far as I know) when you want to share the spindles / array across multiple servers. So if exchange 2007 has managed to reduce substantially the amount of IOp/s it needs you may now need fewer spindles to service it; but whether you choose to provide those spindles via SAN or DAS is likely to be driven by other factors, just as it was previously. I guess the difference is that if you have fewer spindles now you may have less wasted space, but using the wasted space was not really an argument for SAN in the first place – because doing so would just increase the IO load further!


  2. Posted August 21, 2008 at 6:22 AM | Permalink

    Quite. That’s essentially the point. There were plenty of customers who were using SANs but were unable to gain the expected economies of scale by sharing them across several Exchange store servers. That’s because the I/O requirements were so high.

    Now, given enough RAM, the I/O requirements are less, which can mean that DAS is appropriate. Probably using SAS (just to add abother TLA). Whether the DAS spindles are managed as hardware RAID — with an array controller — or software RAID/JBOD — e.g., Windows’ Dynamic Disk — is another consideration again.

    Thanks for the comment; I made a minor edit to the third graf to tidy the meaning.

Post a comment

You must be logged in to post a comment. To comment, first join our community.