none
HPC and Azure - too much storage transactions RRS feed

  • Question

  • I've been doing some tests working with Windows HPC Server 2008 R2 SP1 and Windows Azure (the burst scenario), and I've noticed something a bit odd going on in the Azure storage used by the cluster – there seems to be tens of thousands of storage transactions during the time the Azure compute nodes are up, even when they are not running any job.

     

    For example - I had one large instance (4 cores) turned on for 3 hours, without any job running, and then shut it down. During those 3 hours there have been 128000 storage transactions, according to the daily report. I tried it again yesterday and turned on a large instance for half and hour, and got about 20000 transactions.

     

    I understand there are 2 proxies running in the hosted service (2 small instances), but I haven't seen anything special written to the storage, other than a heartbeat table which was updated every 30 seconds with the state of the large instance.

     

    What's going on? why are there so many transactions? which part of the cluster is doing this? the head node? the proxies? the large instance?

     

    I'm currently using my MSDN premium account that gives me 1M transactions per month, and I've already exceeded this after having a large instance turned on for 12 hours (again, without running any jobs).


    Please mark posts as answers/helpful if it answers your question. Senior Consultant on WCF, ASP.NET, Siverlight, and Entity Framework. Author of Microsoft's Official WCF 4 Course. Co-Author of the book "ASP.NET 4 Programmer's Reference" (ISBN 978-0470505458).
    Sunday, April 3, 2011 7:04 AM