CloudByte releases ElastiStor 2.0
We are announcing the new release of our flagship product Elastistor, the ElastiStor 2.0. The team CloudByte is thrilled to release this technology intensive major upgrade in quick succession to the recent open source project launch OpenEBS. ElastiStor 2.0 brings significant technology additions to our customers and partners, enables new use cases in backup and archival area and broadens the hardware support.
Last couple of years has seen a steady increase in the adoption of Software Defined Storage in the infrastructure space. While working with some of our largest customers, we picked up a lot of skills in tuning the storage for capacity, performance, scale and reliability. Much of this learning has gone into the ElastiStor 2.0 Machine Learning (ML) modules. Many improvements to the new upgrade come in the form of auto tuning the system parameters to provide reliable performance and to ascertain the QoS guarantees.
So, what does ElastiStor 2.0 have in store that makes it that much interesting? Apart from the consolidation of bug fixes from the customers’ feedback and a cool new administrative UI, it has the following major improvements.
- Full-fledged support for backup and archival
- Better Hardware Support
- More reliability
- Superior performance
Full-fledged support for backup and archival
Backup and archival is one new area that CloudByte is aggressively expanding in the product fit and support. As the data explosion continues to happen, the demand for the economical and reliable storage solutions is also increasing. Add to this, the need for hardware refresh forces the movement of existing data large from the legacy EMC, NetAPP systems to the modern SDS systems.
ElastiStor 2.0 brings the following new features for positioning in backup and archival use cases
ElastiStor 2.0 is the first ZFS based storage appliance that provides constant rebuild time when a disk fails. Yes. We added significant engineering effort to re-architect some of the assumptions in the legacy ZFS Pool metadata architecture to achieve this feat. Adding significant technology additions to ZFS without compromising the underlying stability and reliability that ZFS offers is not easy. That is why this feature deserves a special mention. There is a detailed blog on this, but here is a quick summary
The architectural issue in ZFS is, it scrolls through the entire pool metadata for disk rebuild. This makes the rebuild time to be directly proportional to the size of the pool metadata. So, the rebuild time when 500 disks are present in the ZFS pool can be 50x compared to the rebuild time when 10 disks are present in the ZFS pool. This is definitely not desirable and a huge problem for considering ZFS for pools beyond a petabyte capacity especially with slower NLSAS disks.
ElastiStor 2.0 resolves this issue by keeping the metadata on a per-RAID group level and hence making the rebuild time constant. With this, the capacity of a ZFS pool is only limited by the practical constraints such as SAS connectivity and Rack/Power dynamics. With De-duplication and compression enabled, a single ElastiStor cluster can scale-up to 6–8 Petabytes easily.
New All Flash Appliances and Wider support for Hardware
ElastiStor 2.0 now supports installing the ElastiStor OS on ZFS itself. The feature is traditionally known as “root-on-zfs”. With this feature, customers do not need to depend on the hardware RAID cards anymore. An enterprise grade X86 server platform is more likely to support ElastiStor OS.
We have also upgraded our Flash appliances (ESA-A100 and above) with few key hardware changes. They are now populated with Micron S600DC family of SAS SSDs and the disk shelves come in dual 12 Gbps SAS back plane. The SAS back planes are fully redundant and do not need the traditional SAS switches. The FX4 appliance carries 4TB SAS drives and is expandable up to 1 petabyte of usable flash storage within a single cluster. With de-duplication and compression it can exceed beyond 2 petabytes.
Improvements to High Availability (HA) and QoS SLAs
Being highly available is extremely important in Enterprise storage. Disks are the most common components we see issues with. Usually when a disk fails, it gets replaced with the hot spare and rebuild the data. But when you are offering QoS SLAs, a more difficult problem with disks comes onto the horizon. The disks go partially bad and the latency from the disk shoots up. While a disk may still be operational, the latency could increase to higher levels such as 50 ms instead of 5 ms. The legacy robust and reliable zfs stack does not have much to offer in this context. In such scenarios, we have observed the resulting latency of the zvols or zfs volumes increasing proportionally. ElastiStor 2.0 resolves this issue by keeping a track of mean latency expectation of the disk and taking appropriate action when the latency falls outside the expected zone.
Another key issue with ZFS pools we have seen the last few years is the difficulty of managing the time expectation of ZFS pool export and import. The ZFS pool export and import times constitute a major part of the overall HA fail-over and fail-back times. ElastiStor 2.0 has significant improvements to the underlying ZFS to keep the export and import times constant irrespective of the amount of traffic, dirty data, state of the disks etc.
Superior performance
This new firmware boasts of many software enhancements to the way CPU cycles are managed.. in more technical terms how the locks are managed
We have put the 2.0 firmware on the anvil and stressed with more than 20,000 connections per individual volume, saturating the underlying SAS bus bandwidth and achieving 250K IOPS per system node. That is quite an improvement compared to many ZFS peers in the industry and also to ElastiStor 1.x version
We continue to give the free license to get started with ElastiStor 2.0 also. Take a look at the new version of our product right away. Download it now!!!
