Sandia National Laboratories
Rational and Plan for the Red Storm IO Risk Mitigation Project
Presentation to Cray, August 2006
Phase 1 Test Results

Red Storm IO Performance Analysis, presented at the IEEE International Conference on Cluster Computing (paper) (presentation)


Red Storm IO Performance Analysis


The initial charter which instigated the activities accessible through this portal was as follows:

  1. Identify major IO performance bottlenecks when accessing the Lustre file-system on Red Storm.
  2. Propose follow-on project plan to resolve the bottlenecks.
  3. Generalize the performance analysis methodology for subsequent use.
  4. Consider applicability of other Sandia IO initiatives as work progresses.

In response to this initial charter the "Rationale and Plan for the Red Storm IO Risk Mitigation Project" was drafted. A timeline of activities can be found here. A brief description of each test follows, along with a link to detailed information in graph and textual form.

TEST1 - Single client, file-per-process, increasing stripe(transfer) size. 
TEST2 - Increasing client, file-per-process, 1 stripe.
TEST3 - Increasing client, file-per-process, 2 stripe.
TEST4 - Increasing client, file-per-process, 2 stripe, limited OST's, 60:1 oversubscription.
TEST5 - Increasing client, shared-file-access, 8 stripe, limited OST's, 60:1 oversubscription.

James H. Laros III

Related Links
Cray Inc.

Cluster File Systems (CFS)