A glitch in the settlement system meant that trades in the Asia-Pacific region were delayed for 24 hours.
It marked the first major mishap for the global settlement system, which was launched last September and clears about $700bn (£450bn) worth of trades every day. The incident is also an embarrassment for IBM, which runs CLS's IT systems and is the bank's main technology supplier.
A spokesman for CLS Bank said, "We ran out of disc space running the processing systems. This slowed down the processing before the settlement cycle was due to run from midnight."
The settlement bank's databases, supplied by IBM, became overloaded with information and the mainframe machines could not extract or input data, causing the system to freeze. The outage lasted for about three hours.
"It would appear that some of the maintenance runs [which clean up the application for the next day's batch of settlements] did not clear from the database. A huge amount of data was left in," the CLS spokesman said.
CLS said IBM would not have to pay a penalty fee in compensation for the fault, adding that it has installed new safeguards in the system since the incident. IBM declined to comment.
CLS's system took five years to develop and was one of the largest IT projects in financial services. CLS Bank comprises an industry consortium of 67 leading banks, including Goldman Sachs and Citibank.
The CLS service aims to provide a simultaneous settlement service for global foreign exchange transactions between banks by slashing the time taken to settle currency transactions from days to hours.
IBM developed the CLS technology platform using Unix server clusters.
The two mirrored sites and a third back-up implementation are based on clusters of IBM pSeries servers running IBM Unix.
The system runs IBM DB2 databases and Shark enterprise storage servers. Middleware has been adapted from IBM's Websphere product, with Tivoli software responsible for systems management.