NCSA dramatically increases mass storage

by Stephanie Drake

It's been an exciting fall for anyone involved with mass storage at NCSA. Now NCSA computer users are beginning to enjoy the results of the center's mass storage improvements and advancements.

In September NCSA began installing an AML/2 tape library from Advanced Digital Information Corporation (ADIC). Two key features of this system are particularly important to NCSA staff and computer users. First, AML/2 is a scalable library that can grow to eventually hold petabytes, even exabytes, of data. Second, AMLs—automated mixed-media libraries—can handle different types of storage media. NCSA's new AML/2 library can store and mount over 15 different types of storage media—from the smallest DVDs to the largest Sony DTF tape products.

Although the floor space used by the AML/2 is similar to that used by NCSA's old StorageTek PowderHorn library, the AML fills twice as much cubic space—208 cubic feet—as the PowderHorn's 104 cubic feet. More importantly, when completely filled with media, the AML/2—as currently configured—will hold 600 terabytes of data. For the time being, until it is phased out in coming months, PowderHorn continues to hold 120 terabytes of data. In sharp contrast, NCSA's three-year-old IBM library held only 10 terabytes of data in 36 cubic feet of space before it was decommissioned in September. The current configuration of the AML/2 holds 11,400 media tapes; PowderHorn's capacity is 6,000 tapes; the IBM's capacity was just 1,200 tapes.

As users' storage needs grow, AML/2 will increase the size of its footprint in the NCSA machine room. The AML/2 will grow longer with the addition of each new unit. Each unit will provide slots for roughly 5,700 more tapes, depending on which media are used in the unit.

An ever-growing need for storage
The AML/2 library has significantly stepped up NCSA's near-line tape storage facilities, as shown by the following graph of NCSA's 2001 growth in library capacity.

The time period covered by this graph is January 1 through November 13, 2001. NCSA's overall system capacity is considerably larger than its library capacity because a second copy of all user data is stored offsite for security. In addition, for quite some time staff had been manually mounting old tapes that were not part of the libraries. Since the inception of NCSA in 1985 and its official opening to users in January 1986, the storage of data has been critical. Early requirements now look miniscule. But huge increases in the power and speed of supercomputing systems have generated ever-growing mountains of data that need to be stored. The following graph shows that NCSA has doubled its storage capacity nearly every year, a trend that will have to continue as NCSA, SDSC, Argonne National Lab, and CalTech implement the NSF TeraGrid project.

New tape drives offer new capabilities
NCSA's AML/2 library currently holds a total of 20 tape drives: 10 IBM 3590 drives and 10 new IBM LTO drives that entered production in early November. PowderHorn currently holds 7 StorageTek 9840 tape drives. Each 3590 and 9840 tape holds 20 gigabytes of data, while each LTO tape holds 100 gigabytes.

The 9840 tape-drive media are very expensive compared to the other two, but this tape technology does fill a valuable niche: it's good for storing small files and for providing fast access to data. NCSA uses 9840 tapes to store files under 512 megabytes. As a mid-load type tape, the 9840 gets to a particular file faster because it has to spin through just half of the tape instead of through the whole tape.

The 3590 tape drives are used for all other files (those larger than 512 megabytes) currently in the center's mass storage system. The LTO tape drives offer much larger capacity but somewhat slower data transfer rates. Six of these tapes are being used to store a second copy of all data in the mass storage system. NCSA's centerwide backup system, which also writes about 1 terabyte of data, uses 4 of the new LTO drives. In the past, all tapes for the centerwide backup system were mounted manually. It's a nice change to have this system self-contained and automatically mounted in the AML/2 library.

Installation
NCSA staff recorded the installation process and posted the their photos and text at http://spongebob.ncsa.uiuc.edu/adic/install1.htm. It's fascinating to see the pieces of the system come together.

The AML/2 was delivered over several days in early September, and installation began immediately. Drives were installed the following week, and NCSA's Storage Enabling Technologies staff began testing software. During the last week of September the library was being used for production tape requirements. Once the kinks were worked out of the system (kinks such as dropping a few tapes a day!), all has been running smoothly. Technical Program Manager Michelle Butler reports, "We have been happy as clams!"

The new LTO tape drives entered production the last week of October. After a week or so of improving the driver, staff have been very pleased with the performance of this system. For at least two months, NCSA will be testing how the new LTO drives store a second copy of all data in NCSA's mass storage system. After that test period, LTO drives will be used to store large user files in both copies of all user data (the second copy is stored offsite in a different building). Additional tape drives will be needed to accomplish that task. NCSA is preparing to purchase 10 additional LTO drives, 8 of which will be used to store files of user data.

Requirements for mass storage are demanding. Now—with a big range of choices for media storage and a library that scales as large as needed—NCSA has a system in place that can grow to meet these requirements. Still more changes will be coming soon when a new storage area network (SAN) environment arrives. Stay tuned for more about how NCSA is better serving the data needs of our user community.