OIT Service Level Agreement regarding user-sponsored BDUC upgrades ================================================================== by Harry Mangalam v1.1, August 18, 2010 :icons: //Harry Mangalam mailto:harry.mangalam@uci.edu[harry.mangalam@uci.edu] // this file is converted to the HTML via the command: // export fileroot="/home/hjm/nacs/BDUC_ServiceLevelAgreement"; asciidoc -a numbered ${fileroot}.txt; scp ${fileroot}.[ht]* moo:~/public_html; .This documents addresses the case when a user or group wishes to: **************************************************************** . provide a hardware upgrade to an existing BDUC node . add a new node to BDUC . use BDUC storage in either above case - acknowledgement that BDUC is not backed up. **************************************************************** Hardware Upgrades ----------------- The user agrees to pay directly for the additional hardware (CPUs, RAM, disks, add-in cards), which will be added to an existing, working BDUC node. (ie you buy the hardware and deliver it to us or we specify the hardware and charge it to your grant number). - The BDUC nodes have been obtained secondhand from Broadcom and are not guaranteed to have 'any' working lifetime beyond working at the time of the hardware addition. Should the node fail at any time, OIT will return the hardware to the donating user or OIT can attempt to install it in another similar node. In any event, OIT does not guarantee that transferring it will work, nor can OIT support repeated attempts to resusitate the hardware combination. In essence, OIT will make a 'good faith best effort' to maintain the hardware, but can decide at any point to return the hardware to the purchaser. - Further, because of the way that the BDUC cluster operates under the control of an optimizing scheduler, your jobs may not go 'on-cpu' immediately if there is a job currently running. The queue that controls your job will have a priority that will make this infrequent but it can happen. New Hardware ------------ Should a user wish to add completely new hardware to BDUC (if MPC and GreenPlanet clusters are inappropriate), she can do so with the following provisos: - that she consults with the BDUC admin BEFORE PURCHASE to make sure that the hardware is acceptable to the admin (to prevent a hardware and power misconfigurations, correct assumptions about performance, assure rack space and networking) - that she accepts that the node will be operated as a normal BDUC node and that while she can have priority on the node, she will not have complete domain over the node. Other jobs will run on the node at lower priority and she may have to wait for her jobs to go 'on-cpu'. The priority that she has on her node will be set via during the initial consultation with the BDUC admin. BDUC storage is not backed up ----------------------------- All BDUC users acknowledge that *BDUC IS NOT BACKED UP*. The common cluster storage on BDUC is on a http://en.wikipedia.org/wiki/RAID_5#RAID_6[RAID6] (2 parity disks, meaning that the data can survive the simultaneous failure of 2 disks). While this implies a certain robustness, it is not a replacement for data backup *WHICH BDUC DOES NOT DO*. *IF THE BDUC RAID STORAGE FAILS, ALL DATA STORED THERE WILL BE LOST.* So back up your own data. Alternative Approaches ---------------------- If these constraints are not attractive, then OIT can help you find new equipment that will meet those requirements. It will almost certainly be more expensive, but it will come with a guarantee of longevity, replacement rules, and simple exclusivity.