MySpace Replaces Storage with Solid-State Drive Technology in 150 Standard Load Servers
MySpace and Fusion-io recently announced they are working together to reduce costs associated with datacenter operations. Using Fusion-io's ioDrive solid-state disks MySpace replaced 150 of their standard load servers, and reduced the headcount of their heavy load servers from 80 to 30. Overall the companies report in a white-paper published recently a reduction of 51% in server footprint as a result. MySpace, according to the white-paper, plans to replace over 1700 of their remaining 2U servers as they reach their end-of-life.
In a March 2009 press release, Fusion-io said that their most recent Solid-State Drives (SSDs) were benchmarked at a sustained I/O bandwidth of 1.5 Gbytes per second, with 186,000 read I/Os Per Second (IOPS) and 167,000 write IOPS achieved for its 320GB unit- by comparison, a low-end SATA Hard Disk Drive (HDD) is capable of about 70 random IOPS, while high-end SAS HDD models can achieve 300 random IOPS. The unit is built as a Peripheral Component Interconnect Express (PCIe) card, with latency under 50 microseconds, as compared to an HDD with a 4.5 millisecond, or 4500 microsecond, average access time.
SSDs have been evolving rapidly over the last few years, and now appear to be reaching a point where they can be better than HDDs for some applications. The key benefits that SSDs provide are:
- Lower power utilization - An SSD in a server uses about 3 watts of power, whereas a similar HDD uses from 6 to 10 watts
- Faster seek times - As described above, SSDs have seek times which are at least a full order of magnitude faster than HDDs for random access, with SSDs having about 100 microsecond seek time whereas HDDs average around 4500 microseconds.
- Higher reliability - The lack of moving parts means that SSDs are much less likely to have hardware failures than HDDs are - the published Mean Time Between Failures (MTBF) for an SSD is about 2 million hours, whereas for an HDD it's 6 times less, at 300,000 hours.
- Silent operation - SSDs are completely silent when they are running, HDDs are most definitely not
The biggest hurdle preventing SSD adoption, however, is price - SSD storage of 64 Gbytes is still about 4x the cost of HDD storage of 64 Gbytes. However, Jon Stokes of Ars Technica points out that, as cooling and power utilization at datacenters becomes a greater concern, the lower operating cost of SSDs relative to HDDs causes the overall price of either solution to be much closer. Stokes also warns:
Not every datacenter will be able to replicate MySpace's costly feat of hardware compression, though. As a social networking platform, MySpace is essentially a giant database application with a huge number of concurrent connections. The MySpace engineers claim that the ioDrive-hosted parts of the database ran as fast as if they had been cached in a RAM drive. As one of the participants in the aforementioned SSD discussion remarked, even at current prices, SSD already makes plenty of sense for datacenters that are performance-constrained instead of capacity-constrained, and MySpace appears to fall into the performance-constrained camp.
James Hamilton, from the Amazon Web Services Team, argues that power savings alone is not enough to justify the use of SSD. He believes SSD is a great technology to replace hard disks in very high IOPS workloads, or when data cannot go to disk and must be held by main-memory (RAM) caches. Since main-memory is power-intensive and still expensive relative to hard disk ahd SSD, adopting SSD for such workloads makes a lot of sense. But not all data needs SSD, he reasons in his article Replacing ALL disk with SSD?:
But in every company I've ever worked or visited, the vast majority of the persistent disk resident data is cold. Security and audit logs, backups, boot disks, archival copies, debugging information, rarely accessed large objects. Putting cold data without extremely aggressive access time SLAs on flash just doesn't make sense. These data are capacity bound rather than IOPS bound and flash is an expensive way to get capacity.
Whether total or partial replacement for hard disk storage, SSD technology is a good tool in the datacenter operations arsenal. Its superior throughput and reliability makes it an ideal hard disk replacement for high IO loads, while one should reasonably expect the price per unit of storage to reduce as the technology evolves and matures.