Title: Solving Large-Scale Semidefinite Programs in Parallel

Abstract: We describe a portable, parallel and distributed implementation
of the spectral bundle method for semidefinite programs (SDP).
Subgradients are computed via a Lanczos method with implicit restarts, but
specially tailored to handle the block-diagonal structure which occurs
often in applications.  Based on the Message Passing Interface, this code
exploits parallelism to solve (to low accuracy) problems larger than
previously possible.  We discuss various computational aspects and present
numerical results on some of the largest SDP's solved to date.