Craig Ruff, a software engineer in SCD's Mass Storage Systems Group and project lead for the Storage Manager update, peers into diamond, the new SGI Origin 3900.

SGI's Dan Honstein configures a 12-CPU SGI Origin 3900, the newest component of the MSS Storage Manager. The machine is scalable to handle future increases in the amount of data being stored and moved about.

Photos: Lynda Lester, NCAR/SCD

MSS: More data, faster access

The powerful new machine will provide fibre channel access to the MSS robotic tape library and disk cache, increasing the amount of data that can be stored and dramatically improving response time for users.

New tapes

The upgrade will make it possible for SCD to replace old StorageTek 9940-A tapes with high-capacity STK 9940-B tapes that hold three times the data. The newer tape drives, which use fibre channel, deliver data up to three times faster than the old ones, which use an outdated HIPPI channel interface.

The Storage Manager has been writing files up to 50 megabytes to the RAID disk cache since Fall 2003. In the Spring of 2004, SCD began migrating these files to 9940-B tapes and copying user files of all sizes from the older to the newer tapes. This has already resulted in faster read/write times for users.

"The bottom line for users is that when we start moving larger files to the 9940-Bs, they'll see faster access to the MSS," says John Merrill, head of SCD's MSS Group. "Already we see an improvement for files under 50 megabytes. The next step is to add larger files, which we'll be able to do when the new Storage Manager is up and running production, and additional hardware components, including more 9940-B tape drives, have been installed. The ultimate goal is to put all user files on 9940-B tapes, regardless of size."

Bigger disk cache

The Storage Manager also drives the fibre-channel MSS disk cache, which consists of a StorageTek D-178 Redundant Array of Independent Disks (RAID) with a configured storage capacity of 8.3 terabytes. The disk cache is the buffer between the user's computer and the robotic tape library, a "holding tank" where MSS files reside for up to 30 days after they are written. A file can be read from or written to disk faster than to or from tape, which may require a robotic tape mount.

The disk cache itself offers a significant speedup in MSS response time over the old MSS disk farm, which was replaced two years ago. SCD is expanding the capacity of the disk cache to 60 terabytes — a whopping increase over the disk farm's 180-gigabytes. This is significant, because the more files that can be held on the disk cache, the quicker the turnaround time.

"Moving to the higher-capacity 9940-B tapes and expanding the disk cache put an increased load on the MSS server," says John. "Diamond is a more powerful machine and has considerably more functionality than the two machines it replaces — more interfaces, more memory, more CPUs."

The bottom line is that both storage capacity and access speed for the MSS will increase substantially. This will be good news for the NCAR community, which depends on MSS data archives to carry out its important research.

— Lynda Lester