On Fri, Sep 26, 2008 at 1:54 AM, Aaron Blew <[EMAIL PROTECTED]> wrote:
> Hi Michael, > Overall and most of the time, SANs are a good thing. They have several > advantaged over dedicated directly attached storage arrays: > 1.) They're generally a lot smarter about how and when they write and read > to the disks. Often they understand what's going on down at the head > level, > and can leverage that for better performance. > 2.) They've generally got a lot more cache than a directly attached array > (some systems can have up to 256GB of cache) > 3.) They're a lot more reliable than many direct attached arrays. There > have been many many hours put into algorithms to detect and predict disk > failures by these SAN vendors, and they're designed to keep that data > online > as much as possible as their reputation rides on their availabity. Hitachi > Data Systems (as one example) even offers configurations with a 100% data > availability guarantee (so long as the unit has power) > 4.) Having all those spindles under one management/virtualization framework > makes you a lot more agile with how you can make use of your storage. The > MySQL workloads your environment has may not all be striped across all the > spindles within the SANs, segregating the workloads. However, using all > the > spindles available can have advantages in some workloads as well, since not > all databases will be hammering down to the spindle all the time. > > A SAN environment isn't always a trivial thing to operate, but it will save > a lot of time over managing 100s of direct attached arrays and can offer > performance capabilities way beyond what can be practically achieved by > using direct attached storage. > > -Aaron > > > On Thu, Sep 25, 2008 at 6:38 PM, Michael Dykman <[EMAIL PROTECTED]> wrote: > > > Hello all, > > > > I recent started employment with a company which has a lot of mysql > > servers (100+ is my best estimate so far) and have all of their > > database servers, masters and slaves alike, using one of 2 SANs for > > data storage. They servers are connected to dedicated switches with > > fibre to to SANs and the SANs themselves seem to be well configured > > and tuned. > > > > However, it seems preposterous to me that all those very busy > > databases should, by design, have a common bottleneck and share a > > single point of failure. I am not deeply knowledgeable about SANs or > > their performance characteristics; my reaction thus far is pretty much > > intuition but I help can't but picture the simple analogue of single > > disk or a RAID 10 with synchronized spindles frantically thrashing > > back and forth to respond to tens of thousands of queries per second. > > > > Would anyone care to comment? Is my concern justified or am I merely > > confused? > I can't comment on the details, but I know our large medically based institution uses a SAN, and the transition to it was well thought out and implemented. To my knowledge, client applications have always been the cause of downtime, not bottlenecks on the SAN.
