Jay

Jay is a special high-end analysis and visualization system installed especially to enhance the effectiveness of our new core resource, Itasca, and for efficient transfers of large data sets. It has fast processors and large memory coupled with large local storage. The system is connected to Itasca and to the campus backbone using high bandwidth networks for faster data transfer. Access is expected to be primarily through advanced reservations explained below.

System

Jay is a HP ProLiant ML370 G6 server with a pair of 2.66 GHz Xeon X5550 quad-core chips and 64 GB of RAM.

Home Directories and Local Disks

Jay mounts MSI's standard /home directory, which is the same as on all other SDVL systems. Jay also has 71 TB of locally attached SAS-based storage mounted as /scratch.sas. This disk will be scrubbed at the beginning of each reservation, and, hence, users need to transfer their data to more permanent home or project space before the end of their reserved access  (see Reservation and Queuing Policies for more detail).

Networks

Jay's default network connection is a 10 Gb Ethernet interface directly to MSI's core network, which is tied via 10 GbE to the campus core network. This facilitates high-bandwidth data transfers in and out of MSI using Jay's large local disk as an intermediate storage pool. Jay also has a quad-data rate (QDR) infiniband connection to Itasca, our new 8000+ core machine. This link allows data to be transferred to or from Jay to Itasca's /scratch file system.

To access Jay follow these steps:

Jay does not allow regular ssh sessions. Use of jay is controlled through reservations and queues. If jay is available a limited interactive session may be requested using: 

  qsub -I -q jay

which provides a 30 minute reservation. a longer interactive reservation can be made(up to 12 hours) using:

 qsub -I -q jay -l walltime=hh:mm:ss

Once a reservation is setup you may ssh or login to the jay console for the duration of your reservation period.

Reservation and Queuing Policies

 

While Jay can be accessed when not already reserved by someone else,  the best way to ensure that you have clear access to Jay is by making a reservation in advance The reservation can be no less than 24 hours and can extend up to a maximum of 2 weeks. The reservation allocates the full machine for your use for the time requested. During that time period, you can submit or interactively perform multiple jobs that could include computation, visualization, and data transfer. You can request only one reservation at time, and your next reservation can be made only after the existing one is completed.  However, you can request an extension of an ongoing reservation, provided it does not conflict with another pre-existing reservation . To request an extension please contact us at at least one working day before the end of your open reservation, in order to allow the reservation end to be reset.

 

Other jobs

 

If you do not have a reservation, you can submit jobs to Jay’s queue with a maximum wall clock time of 12 hours. But remember that jobs submitted during a reservation get higher priority than jobs submitted otherwise. In addition, any data stored on Jay's local disk will be scrubbed when a reserved session starts. Due to these restrictions, you might find other queues to be more efficient in running your jobs. You can check the reservation schedule as noted below.

 

Setting Up a Reservation

To review Jay's reservation schedule use this Jay Calendar.

To make an advanced reservation use the reserve-jay tool available now on SDVL workstations. reserve-jay is very similar to the rset tool available on the core resources.

to make a reservation for 5pm on Nov the 14th for 1 hour:

  reserve-jay -s 17:00:00/11/14/2010 -d 1:00:00

reserve-jay now also supports some limited natural language processing:

  reserve-jay -s "next monday at 5pm" -d 1:00:00