NERSC 1998 User Survey Results

These are the complete responses to this question, with a few minor edits. Some responses have been repeated (to group them under multiple categories).

T3E Comments:

• I am mostly using the T3E machines. It turned out my experience on it is a very good one.

• I am very greatful of the opportunity of computing at NERSC's T3E. However, our allocation for 1998-99 is very small.

• The T3E is now our main computing resource, so our ability to compute on it has an overwhelming influence on our impressions of NERSC.

• T3Es response time is great (I guess nobody's using them)

• T3E queue policy needs improvement. We need longer runs.

• It has been very difficult to obtain adequate job turnaround on the T3E during the last four months of FY98.

• The Network access is about the only thing I have a problem with concerning NERSC. It seems to be a horrible shame that ssh can not be installed properly on these machines. The fact that it wasn't has caused me some grief this past year.

• Much time was wasted (both cpu and my hours) trying to "debug" a problem with Cray's temporary memory allocations during execution at the source level (invisible).

• We are carrying out a vary large simulations project on the T3Es. The queuing system is extremely awkward from our point of view, but we have communicated this clearly to Brent Draney et al and there does not appear to be a short term solution.

• T3E: the machines are of course fine -- they are great machines [mcurie and pierre] - I appreciate very much that mcurie has the 256 memory per node. however: mcurie is so loaded with jobs, that the throughput of the jobs is quite slow. though nersc has pierre for development work, I do not use it for that purpose, since test jobs are too long in the machine to make code development efficient. ==> I love the t3e -- but I don't like the way you set it up.

• The overall rating for NERSC is a little low for the following reason: We can't really make use of the T3E since the FORTRAN code we are using has a lot of VMS extensions and is to a large extend legacy code. This make it very hard for us to port our software to the T3E, we finally gave up since the time investment was unaffordable.

• My only complaint is that the system is too busy. [C90 user]

• C90 response time is very slow.

• The turning time for large jobs are very slow, especially for the 512Mb queue which can not be done locally.

• Really haven't used the T3E, and my use of the J90s is limited because the C90 seems faster with small jobs.

• The J90 performance compared to our local workstations is inferior. We will miss the C90 but understand that it must go. We hope that the J90++ is a lot better.

• Biggest problem is constant changes in system so programs that were running stop doing so. Find a way to make changes more transparent

• I don't understand the J90's directory names.

• Most of my work is done on the C90 cluster and running batch jobs. I don't use more than half of the services you list.

• CFS user interface remains brain-dead.

• I don't even know what "HPSS" is, so again I can't comment.

• Disk space management is a fiasco.

• We are still using CFS, rather than HPSS, mainly because we haven't had any real information about HPSS. (I've looked at the web pages, but having to go through the effort of getting passwords, etc. there is not much incentive. This is especially true since I then have to teach other people what to do.)

• Network access has improved recently, but we still suffer from periods where the bandwidth is too low to work with.

• In general I am very pleased to have access to the NERSC facilities.

• The use of collaborative tools to help interaction between users and users and nersc staff would be fundamentally valuable.

• I find that the nersc consulting group and the web pages have been extremely helpful to me in completing my work on the t3e.

• Consulting services are very good.

• I have found consulting services to be prompt and helpful.

• Consultants are helpful, particularly Jonathan Carter and Frank Hale.

• Consulting service is very uneven, depending on who your consultant is.

• Telephone consultants are excellents. They walk you through all your problems and attempt to teach you more about the system. Very friendly and patient.

• I am sort of the second line for users in our area - I have always found the consultants and account services people willing to listen and lend a helping hand.

• The consultants in nersc have been very helpful and very responsive.

• For me, at present, NERSC is access to the C90 and CFS and contact with your consultants. With all of these I am very well satisfied. I especially appreciate the helpfulness of your consultants.

• Consultants are quite good and generally friendly.

• The consulting staff are very prompt in responding to user's questions and comments.

• Online documentation probably should better cover things not available through man pages.

• New Users should be given batch templates to aid the tutorial process.

• I find that the nersc consulting group and the web pages have been extremely helpful to me in completing my work on the t3e.

• Technical info is sometimes hard to read/understand.

• It is sometimes hard to find information on the web site, especially info needed to submit runs. For example, it took me awhile to find the maximum memory sizes for various queues and also looking for sample commands or batch files for running jobs I think the technical information and online help could be better organized.

• It takes more than a week to set up new accounts which is too slow.

• On one occasion, it took several days, longer than I felt should be necessary, to have an HPSS password assigned for a collaborator. Otherwise, all my requests have been taken care of promptly.

• when one types setcub view, one should see the percentage of time left in the repo

• Allocations were assigned monthly, this was not convenient.

• The allocation process & account services (creation, password changes) are quite cumbersome. Because of security concerns, I (the user) cannot initiate a request to NERSC. Rather I have to contact our "official representative" who then transmits my request to support services. If that person is on vacation or otherwise not reachable, I am up-the-creek. I wish there were a better way!

• I am sort of the second line for users in our area - I have always found the consultants and account services people willing to listen and lend a helping hand.

• The javascript interface to the allocation process is still rather overcomplicated to use.

• Account allocations and quotas have seemed to be a bit rigid and inflexible. Unfortunately that has sometimes forced me to put time into finding work arounds to administrative strictures which cost myself time and my project efficiently.

• The structure of the queue's and the method of charging on the C90 makes the amount of the typical allocation (at least my typical allocation) useless. It is impossible to get a reasonable turnaround time in any queue except the high priority ones, and with these a single run kills the entire allocation. Either allocations need to be higher, or the queue structure needs to change to make them more usable. I guess the other alternative is to estimate the required CRU in the allocation process using the highest priority queues, however I suspect this would not be acceptable in the allocation request process.

• Allocation Process: I would prefer the allocation process to be more flexible, i.e. more than once a year. it is sometimes difficult to estimate the usage over the entire year. why isn't it possible for smaller allocations to apply e.g. quarterly - that gives a possibility to ask for some extra time if needed. I also would like to know the evaluation criteria for the proposals.

• The allocations process seems to be a ``black box.'' Frankly, I have no idea what real criteria are used to set allocations aside from obvious things like Grand Challenge allocations. Since I am on a GC and a PI on another project, I have a split perspective on this issue. From the GC point of view, everything is great. However, on the other project, the allocation process has not been fair (OK, that's my view of course) apparently because the project is in a new field without an entrenched supporting ``old boy'' network, and hence slips through the cracks --