Han o Sinox's Advanced file system ZFS easily beats Microsoft refs and Linux for Btrfs

SINOX2014 introduced ZFS advanced file system to easily beat Microsoft's refs
Sinox the power of open source is already gaining a leading edge in file systems for high-performance storage servers.
This fact tells people that Windows is not invincible.

File system rivalry: Microsoft refs VS ZFS

ThunderSource: It expert network2013-07-30 08:47

　　The advantages of "IT specialist network storage" storage are generally considered to be highly reliable for hardware-driven, RAID-based file systems. The truth is that significant potential errors persist, especially as the system expands for cloud architectures and big data applications. This potential error, due to the lack of awareness of the operating system or hypervisor low-level data errors, the abstraction of storage resources through the stack, data integrity is assumed to be trustworthy.

Also true, simple raid parity is insufficient to accommodate thousands of disk volumes. If you want to learn more about how the storage methodology has frustrated the cloud revolution, see Professor David S. Rosenthal of Stanford University's data storage paper "Keeping Bits safe:how hard Can It is?".

With the Times, the storage delivery engine needs to become smarter and more transparent to the rest of the stack, which means better file systems, the ability to detect and correct the wrong volumes in real time, and not simply offline routine error detection. Two options are Oracle (Oracle) ZFS and Microsoft's refs (resilient file system, elastic filesystem).

Sun's start to real-time error detection dates back to 2005, at the initial development of Solaris 11 (codenamed "Nevada"). The open source of its prototypes, the next generation of file systems, is known as ZFS, and continues to be developed and improved until it is acquired by Oracle.

Microsoft is trying to create some similar capabilities in its new refs file system. Refs is a feature of Windows Server Storage spaces. At its core, refs tries to solve basic problems like ZFS while maintaining the compatibility of the NTFS file system to support traditional Windows applications, services, and infrastructure.

Both refs and ZFS provide some basic improvements on top of traditional file systems. First, data errors are detected and repaired in real time, without the need to uninstall or interfere with access to the appropriate volume. For ZFS, these are performed at the data-level and meta-data levels, and by using layered checks, the file structure is uploaded to the root node.

When a file is read, the checksum of the read file is compared with the stored checksum, and the error is automatically detected. In refs, the checksum is placed in the file metadata. An additional optional feature called integrity streams ensures that when changes are written, they are written to a different member volume to ensure that the original is not compromised. This error detects anti-bit corruption, degradation of data stored on disk, hardware failures, firmware failures, cosmic rays, and other complete destructive events that do not affect the availability of online file systems.

Although these two file systems have similar characteristics, the performance of ZFS's striped volume calibration stands out. According to Microsoft's documentation, the checksum on refs comes from pairing Microsoft's storage spaces feature and requires that the existing parity information "be read and processed before a possible new write occurs".

Microsoft advocates the use of dedicated SSDs for logs to alleviate poor write performance, and the way ZFS fights this effect is to write parity information directly in a variable-sized stripe. This alleviates the harm of write speed and corrects the notorious "RAID 5 Hole", a limitation in traditional RAID systems that allows data to be lost in the event of a crash or power failure, which in fact constitutes a completely new paradigm for the RAID implementation parity mode.

The cache implementations of the two volumes are significantly different. Refs can accommodate SSD caches, but there are not many features about refs caching, although they do not support DRAM caching.

In contrast, the ZFS cache is well documented, and the DRAM cache is a core feature. ZFS provides three levels of caching: ARC, which is the smart memory cache to RAM, and uses as much available RAM as possible; L2arc, this is a one-time (non-storage) SSD read cache, and Zil, which is SSD write cache, buffers write to the underlying volume.

Location of arc in the entire ZFS file system structure

ZFS supports the use of PCIe solid-state drives or ultra-low-latency battery-powered RAM disk devices as ZIL to significantly reduce write latencies to ZFS volumes. and Microsoft's only relevant statement is that "customers can use third-party solutions for this purpose."

These two file systems support snapshots, encryption, oversized disks, massive files, and the maximum volume capacity of EB. However, refs does not support compression or data deduplication, both of which are core features in ZFS, even though Windows Storage spaces supports data deduplication on NTFS volumes.

The final result, in the non-scrutiny seems to provide similar functional products, in fact, very different. On the face of it, refs and ZFS seem to be rivals, but Microsoft's first-generation refs products are hard to compete with the more mature ZFS. Compared to ZFS, which has developed for more than 12 years, refs still needs to do more to keep Windows competitive.

There are some important limitations to refs, including:

The existing NTFS disk area cannot be converted directly to the refs format, and data can only be moved and copied manually between the two file systems.

Refs cannot act as a boot partition, which means that the server 2012 system must mix NTFS with refs two file systems and start with an NTFS partition.

Refs does not apply to Removable storage media (external hard drives or USB portable disks).

Refs itself does not contain data deduplication, and it cannot be combined with the new deduplication feature in Server 2012 (the server 2012 Deduplication feature applies only to the NTFS file system disk area).

Refs itself has no built-in writable snapshot functionality, but users should be able to create writable snapshots directly from the storage space virtual disk on refs's underlying system through other tool software.

Copy-on-write file system comparison: ReFS vs. ZFS vs. Btrfs

(1) equivalent to Copy-on-wirte.

(2) file and storage services management tools through Windows Server 2012.

(3) Raid Z and Z2 are equivalent to RAID 5 and 6.

　　Small Comments : From the above can be seen, relative to the ntfs,refs provides a lot of improvements. While there are a number of limitations relative to mature zfs,refs, it is invaluable as an innovative product with the Windows Server 2012 operating system. Microsoft is expected to make more improvements in subsequent releases, resulting in better experiences and more choices for the data storage file system.

Han o Sinox's Advanced file system ZFS easily beats Microsoft refs and Linux for Btrfs