Facebook smacks away hardness, sticks MySQL stash on flash
Replaces HDD with Fusion-io flashiness
Magic Quadrant for Enterprise Backup/Recovery
Facebook is using Fusion-io server flash cards in its datacentres to store MySQL data as well as process it faster because it's better than using disk drives.
Piper Jaffray analyst Andrew Nowinski tells us that Facebook is using Fusion-io ioDrive PCIe flash cards for capacity as well as performance acceleration through flash caching:
[Fusion-io] products are now being used for capacity storage purposes … Facebook ordered the new ioDrive2 cards as a replacement for its primary back-end storage, which we believe came at the expense of LSI and their RAID controllers and the corresponding 15K RPM hard disk drives... The pricing of this deal was actually higher than the alternative competitive offering... The performance benefits of the Fusion-io solution greatly exceeded the HDD solution, making the price/performance much more attractive.
Facebook is now "hosting [its] MySQL databases on Fusion-io cards, rather than using an HDD solution from LSI." Nowinski tells us one reason is that MySQL has a journalling system wherein each row of data is written twice, once into the database table and again into the journal for redundancy and protection against a write failure. This uses up storage capacity and lowers performance.
He says: "By leveraging Fusion-io cards and the new SDK, customers can turn off this journalling system, since the Fusion-io cards also have a patented logging system that protects data in the event of power failure. By turning off journalling, management stated that customers can reduce the amount of data that is stored by 50 per cent, increase throughput by 33 per cent and decrease latency by 50 per cent versus the alternative HDD storage."
That would have got Facebook's attention.
Apple signs up
Nowinski also believes: "Apple could actually displace Facebook as the company’s top customer as Fusion-io expands into more capacity-oriented applications beyond iCloud."
The background to Fusion's adoption by Apple was this:
Fusion-io’s footprint at Apple was built by displacing a networked storage vendor in support of their iCloud application. Fusion-io beat Oracle Exadata for this project because their solution offered 4x the performance of Exadata and half the cost, which equates to a 16x cost/benefit advantage. … The significant cost advantages were derived from the consolidation of Oracle software licences that Apple was paying to host their Oracle databases.
Apple deploys the ioDrive cards in twos for redundancy so the cost advantage falls back to 8x, still substantial.
He thinks that Cisco's qualification of ioDrive cards for its UCS servers is progressing well.
Fusion faces competitive FUD flack attack
Nowinski said he had been briefed by Fusion-io management about "FUD" apparently being spread by PCIe Flash card competitors who say Fusion soaks up host CPU cycles by not having controller functions on its cards.
Nowinski tells us:
Many other competitive PCIe solutions, which do not integrate into the host OS, take on processes from the CPU that are needed to store data onto flash memory by emulating a disk drive. It is perceived that by offloading these processes from the CPU, the performance of the CPU improves, since it can then focus on the application that it is running, rather than these disk emulation processes.That is technically true, but the Fusion-io cards do NOT emulate a disk drive. They store data natively on flash memory and do not require a translation to the various storage protocols to storage data on the flash. Therefore, there are no processes from the CPU that have to be offloaded. Fusion-io by-passes disk emulation entirely, thereby driving greater performance. The perception is that since they are not offloading these processes, the Fusion-io cards are somehow bogging down the CPU, which believe is nothing more than inaccurate FUD.
Lastly Nowinski said: "We believe Fusion-io is working on integrating more data management functionality, such as snapshots, replication and deduplication, onto their ioMemory platform."
Piper Jaffray is a Fusion-io supporter and provides investor analysis. This means its comments could be biased by a wish to support its investment thesis. Take that into account when coming to a conclusion about the information we've presented here. ®
Does Fusion-io have a sustainable competitive advantage or will it get blown away by a hurricane of other PCIe flash card vendors attacking the market, such as EMC, Intel, Micron, OCZ, TMS, and many others? Give our Forum readers the benefit of your views please.
COMMENTS
50% space saving is stupid fiction
DB journaling will write the data twice for recovery, but the old journal will be dropped after a DB backup (or mirrored checkpoint), so the journal is never as big as the data… unless you never checkpoint or backup… in which case it will grow to 100%+ (but that’s academic ‘cus a recovery would fail anyway)
33% decrease in latency is also iffy because journal-write has almost no seek latency & a decent SAN will have NV write cache.. SSD will win big on random read.. but the rate depends workload.
For any workload more important than the musings of a spotty teenager, you’d want the journal anyway for replication.
Re: This doesn't make sense
If the database is CPU-bound, TheCreditCruncher's thoughts make perfect sense. But if the database is I/O-bound, switching to a faster storage medium could get more performance out of fewer CPU licenses. Most solid state storage basically makes I/O wait times disappear, so CPU utilization goes up, so overall database performance goes up beyond target metrics, so the number of CPU or server DB licenses required goes down. This doesn't apply to all workloads, of course, but many heavy-hit databases follow this pattern.
Turn off logging
....if you don't care about recovering your DB! FS journaling or any HW equivalent is missing the point. The figures given for space saving when turning off logging are also BS. The log file doesn't grow with the size of the DB!!!
Nothing against Fusion-io but if they really said this they know nothing about databases.
IT infrastructure monitoring strategies
Requirements Checklist for Choosing a Cloud Backup and Recovery Service Provider
Data control in the cloud
Cloud based data management
Enabling efficient data center monitoring
