Discussing all things virtualization and storage in the data center.

Something for everyone

Seeing as how some posters think this is a blog site that does not offer unbiased discussion, I thought I would raise something that all enterprise wanna be storage vendors have to deal with.  So please feel free to put in any discussion that you think is worthwhile especially if you are a marketing person and you would like to tell me just how good your non HDS storage is with something like this.

The organisation that I work for seems to value email above everything excluding the mainframe.  So we are running Hitachi Storage Cluster with Microsoft Exchange and it uses synchronous True Copy to ensure not one single totally worthless joke in an email is lost.  Mind you this also has something to do with Microsoft but I wont point any fingers at them.

I have come to the conclusion that Exchange 2003 is probably the absolute worst application that I have ever seen when sync True Copy is used (apparently as Microsoft does not sanction asynch True Copy).  Exchange seems to want to do random writes during "quite" periods to its data devices which means it tends to do large bursts of data in 4 Kb chuncks.  Due to the fact that HSC relies on MSCS, we are (apparently) limited to basic disks in the servers so host based striping is not available.  So, we use LUSE as Windows (apparently again) needs drive letters to mount disks in a cluster. 

Add all these unfortunate things together and sometimes we get some some poor write response times that obviously has something to do with the two USP's we use (at least that is the view of the Exchange team with quickly spreads across the organisation).  Lets not forget the Brocade 48000 directors and the Cisco DWDM links (8 Gbps all up) in between.  We are talking about serious money here just to run email which sometimes does not work all that well due to poor placement of users and its totally wierd random writes.  Main culprit is queue depth in my opinion as sometimes it goes ballistic and the IOPS on the LUNS head skyward at a high rate.  Once that happens, performance goes pear shaped.  So, I am crucified by the limitations put on me with basic disks and MSCS and the Microsoft consultants who know everything.

The good thing is that Exchange 2007 is out and runs something like a trillion times better than 2003 and also the (oddly named) concatenated parity groups for the USP means that I am not so stuffed up with poor random write performance and LUSE disks anymore.  I did some tests with a 28D + 4P array group and host based striping and believe it or not, the performance improved in the order of up to 6 times on average and about 10 times during peaks.  I must admit it was jaw dropping.  However, I still have the same problem with Microsoft and host based striping as they believe once you get Veritas involved, MS is hands off everything including OS system calls...

So am I the only person in the world that hates Exchange 2003.  I want our organisation to use something like Gmail but that means calenders wont work.  Boo Hooo..  Many millions of dollars of high end infrastructure just for email.. What a joke.

Oh BTW, once we start using Exchange 2007 in production, we don't need True Copy anymore as Exchange does something like log shipping which is bound to work much better as it is async...  Talk about a no win situtation for the True Copies et al of the world.

EMC have asked me to look at Smarts (Incharge I guess).. I wish (oh how I wish) it worked on HDS and IBM equipment.

Stephen

comments powered by Disqus