Aug 102005
Someone recently pointed me to http://www.quickshift.com/Product_QS-SQL.shtml. These guys claim 3x and 4x times performance gains with SQL Server. The product apparently sits between the OS and SQL Server and they improve performance by managing the "flow of data and operations." They don’t give details on the technology up front – you need to contact them first and then they revert. The technology is claimed to be certified by several vendors including Microsoft.
This sounds too good to be true. The only reason I am being sceptical is from what I know about SQL IO, it uses a WAL protocol. WAL requires that log record entries be written before the corresponding data pages. This in turn means that for SQL to work correctly, writes have occur in the order they are sent. Now if QuickShift is using write re-ordering at all, this may not work. SQL IO has a couple of other such requirements also (all given in brief on the link above). These may cause an optimization engine which is unaware of these requirements to fail.
The tool QuickShift have used – SQLIOStress.Exe - ensures rudimentary IO safety. From the link above, "SQLIOStress does not guarantee or warrant data security or integrity. It was designed to provide base line testing of a hardware environment, and it may expose potential data integrity issues." Passing this may not be enough to gurarntee proper working in all scenarios. The only way to find out would be some fairly heavy testing. I would like to test sceanrios like loading the server with trasactions and then cause sudden shutdowns and then recover databases. Such a test would reveal if the intermediate layer holds up.
All said, these guys do have a couple of customers in production so this technology should be holding up. I am hoping that this is actually true – this would be great news for the databases community.
Recent Comments