Shulou(Shulou.com)06/01 Report--

Flash memory is undoubtedly the biggest market bright spot in the storage market in 2013. Flash memory has full vitality at the capital level, market level and technical level, among which the all-flash array market is the most concerned. A large number of start-ups and traditional storage giants have invested and paid attention to all-flash arrays, making the overall market extremely active. All-flash start-ups represented by ViolinMemory and PureStorage are also pinned on new hopes to break the traditional storage market pattern, but traditional storage giants will not sit still. EMC has announced its all-flash array XtreamIO. About a year ago, EMC acquired its all-flash array-related technology through the acquisition of Israeli all-flash start-up XtreamIO. During this year, there were often reports about XtreamIO, but it was always in a state of "not knowing the true face." After a year of integration and planning, EMC finally officially unveiled its all-flash array XtreamIO.

The current flash memory market, especially the all-flash array market, is only just beginning, both in terms of market size and product technology, which are in the early stages of development. For example, the market is still full of all-flash arrays that use traditional storage array architecture and replace mechanical disks only by SSD. Even EMC XtreamIO's all-flash arrays are not perfect, with competitors criticizing UPS power for each node and not yet part of the entire data center hierarchy. However, in the long run, standardized hardware will be the trend, software functionality and architecture will be the key to determining the future competitiveness of all-flash arrays, and EMC XtreamIO will undoubtedly set a new benchmark in the market once it is introduced.

Figure 1: Gartner's 2012 Flash-based Storage Array Revenue Statistics show that the total market revenue for Flash-based Storage Arrays is approximately $370 million. All-flash arrays are still in their infancy and the overall market is small.

EMC believes that flash memory needs to solve more than performance problems. Focusing on hardware and physical performance will put flash memory in a niche market. EMC, as the world's largest storage vendor, is committed to how to use flash technology and software-led architecture to make it affordable for users. It is easier, faster and faster to serve the next generation virtual data center. EMC XtreamIO technical director Ren Yuxiang is even more blunt: Most of XtreamIO's intellectual property patents are concentrated in the software field. XtreamIO's four characteristics of data placement based on 4K data block content, two-level metadata engine, XtreamIO data protection (XDP) and metadata in shared memory enable it to completely get rid of the traditional array thinking mode and architecture.

After the official launch of XtreamIO, EMC also launched an offensive against ViolinMemory, Purestorage, Solidfire, etc., which are popular vendors in the current all-flash field. It boldly threw out product function comparison charts and raised many technical questions to competitors, which also caused a rapid response in the entire all-flash array market. EMC's style of daring to dry goods is far more effective than manufacturers who often claim to spend $1 billion.

XtreamIO All-Flash Array Overview

EMC XtreamIO is an all-flash array with scale-out capability, and each array unit is referred to as an X-Brick. Each X-Brick unit is configured with 25 eMLC SSDs with a capacity of 400Gb, so the raw capacity of an X-Brick is 10TB. EMC claims 7.5TB of usable capacity per X-Brick due to the new architecture, and that's without deduplication optimization.

Figure 2: EMC XtreamIO all-flash array X-Brick module appearance diagram. Each X-Brick is 5U high and consists of Storage Controller, DAE, UPS power supply, 1U dual-socket server per controller, 4 8Gb Fibre Channel ports, 4 10 Gb iSCSI ports, 4 40Gb Infiniband ports, and 2 1 Gb management ports.

Figure 3: Schematic diagram of each X-Brick architecture of XtreamIO, consisting of two storage controllers and 24 eMLC SSDs.

Figure 4: EMC XtreamIO scales out to four clusters (two nodes per X-Brick, four clustered X-Brick means eight nodes), IOPS can reach millions, and Infiniband switches interconnect between X-Brick.

EMC XtreamIO technical director Ren Yuxiang said XtreamIO has achieved eight X-Brick (16 nodes) and even larger array expansion in the lab environment, and EMC will soon provide arrays with eight X-Brick configurations. "In theory, there is no difference between 16 X-Bricks and 32 X-Bricks, but there are a lot of test and real-world scenarios involved, i.e., whether the customer currently needs such a large capacity and performance. "Ren Yuxiang added.

According to EMC, X-Brick features non-destructive XIOS software and hardware upgrades, high-reliability hot-plug upgrades, and no single point of failure. Each X-Brick in a clustered XtremIO array can have different storage capacities; VMware VAAI and vmware multi-channel I/O support are integrated, and data volume can expand the entire system; all metadata storage, configuration, and management are performed in memory, without involving SSDs; and rich data protection features have been integrated with other EMC products.

Once EMC XtreamIO was introduced, it was odd that each X-Brick required UPS power, and two UPS if purchased separately. EMC XtreamIO explains,"Every X-Brick currently requires a UPS to solve power problems. The reason why an X-Brick needs two UPS is because in the case of a single X-Brick, the power supply also needs to be configured with two UPS backups to prevent UPS failure in the case of a single X-Brick. If you purchase more than two X-Bricks, you can configure one UPS for each X-Brick. As for why X-Brick needs UPS, it's because XdreamIO stores all metadata in memory, and metadata operations are performed in memory. Therefore, in the event of any hardware and software problems, memory information will be completely lost, such as a data center power failure, UPS can ensure that there is enough power to write data changes in memory to SSD. We also don't like UPS solutions because batteries are bulky and take up space. We're going to take other approaches in the future, and there will be new technologies to replace UPS on the next generation hardware platform. "

"Content-based data placement" breaks traditional patterns

Content-based data placement refers to EMC XtreawmIO all-flash arrays that divide data into 4KB hash blocks that are completely fragmented and stored not by logical address but by physical location. EMC XtreamIO Technical Director Yuxiang Ren emphasized that 4K block-based content placement is the foundation for all other software functionality across the all-flash array.

Figure 5: EMC XtreamIO's content-based data placement completely breaks with the traditional disk-based data storage model of the past.

In fact, XdreamIO, an array technology based on data fragmentation, has precedents before. Similar arrays such as 3PAR and XIV are based on this strategy. However, traditional disk array data storage is based on logical addresses, and disk fragmentation is inevitable over time, which affects performance. EMC XtreamIO all-flash array strategy is to completely disrupt the data, complete fragmentation, based on 4KB data blocks through fingerprint identification, storage, management, etc. EMC believes that the immediate benefit of this technology is that inline de-duplication becomes very easy.

Figure 6: EMC XtreamIO data placement based on 4KB data blocks, using a two-level metadata engine, enables inline deduplication of all-flash arrays to be easily achieved through two-step metadata mapping.

Figure 7: After data is re-deleted by the core engine, it is evenly distributed among the X-Bricks through the 40Gb high-speed Infiniband internal switching network.

EMC XtreamIO technical director Ren Yuxiang said: "Traditional disk arrays are stored on disks according to the order of logical addresses, and sequential operation performance is the best. If traditional disk arrays do inline duplicate data flash operations, some operation orders will be broken, resulting in disk fragmentation, thus causing array performance to suffer. This is why traditional disk arrays are difficult to do inline deduplication. EMC XtreamIO completely ignores the logical address mode, and decides where to put 4K data according to the fingerprint technology based on the content of 4K data block. This completely fragmented way makes it very easy to realize inline deduplication. If the same 4K data block is written to only one address, when demultiplexing is an inherent function rather than a superposition function, the higher the demultiplexing ratio, the better the write operation performance, because only the pointer needs to be set in the far data. "

It is reported that EMC XtreamIO's fingerprint technology based on 4K data blocks is a hash algorithm. EMC means that by using a good hash algorithm, it will ensure that the data is distributed evenly across all nodes of the structure.

XtreamIO uses a two-level metadata engine. The first step of metadata mapping is logical address to fingerprint mapping; the second step is fingerprint mapping to internal physical address. Once the two-step two-level metadata is completed, the flexibility inside the system is increased. XtreamIO can map to which physical address inside according to the fingerprint. Traditional RAID has a large write amplification, that is, the proportion of write erase operations at the back end and write operations at the front end. Both RAID 5 and RAID6 will amplify write operations a lot, which will result in performance loss for traditional arrays and loss of flash life for all-flash arrays. XtreamIO will greatly reduce the write amplification of traditional RAID. "Ren Yuxiang added.

Figure 8: XtreamIO data protection mechanism XDP, through the two-stage data engine to avoid RAID 5, RAID6 write amplification, to achieve RAID6 protection effect.

Figure 9: EMC XtreamIO XDP principle.

EMC XtreamIO technical director Ren Yuxiang said: "XDP is first of all double backup, and RAID6 allows XtreamIO array colleagues to have two flash disks, if errors can be corrected, will not cause data loss. Colleagues, it also avoids the write amplification that occurs with traditional RAID. "

Shared Memory Metadata-The Secret of XtreamIO High Speed

As mentioned earlier, EMC XtreamIO moves the storage, management, and manipulation of metadata to the memory tier for operations. By storing metadata in memory in this way, shared metadata will no longer be written to SSD; thus reducing the various write operations of flash memory and extending its service life. The way memory metadata is shared also allows many software functions to be done at the memory level.

"Shared memory metadata allows many functions to be accomplished by manipulating metadata," said Yuxiang Ren, EMC XtreamIO technical director. For example, copy function, virtual machine template copied to hundreds of virtual machines, traditional disk mode can only rely on reading and writing all the data blocks, very time-consuming, IO-consuming operations. XtreamIO copy operation is completely in memory, to solve the metadata operation way, the speed will be very fast. "

Figure 10: EMC XtreamIO SnapShot Snapshots can be processed faster through metadata.

Through metadata operations at the memory level, functions such as virtual machine cloning, copying, database cloning, snapshot processing, etc. will greatly improve processing speed. Of course, this approach is controversial. Steve O'Donnell, former CEO of Greenbyte, said,"EMC removed the garbage collection mechanism completely from SSDs, so that locked NAND could not be controlled properly at the system controller level. Steve O'Donnell also pointed out the pitfalls of using an X86 server as a controller, saying,"While putting metadata in memory allows arrays to run at extremely high speeds, controller boards are notoriously unreliable. "This may explain why the X-Brick needs to be configured with UPS power.

XtreamIO: This Year in Integration

After EMC acquired XtreamIO, it did not rush to formally launch a new product, but rather consolidated and integrated all aspects. According to Ren Yuxiang, Xtream leverages EMC to provide the best hardware platform and global supply chain to ensure the quantity and quality of XtreamIO arrays. In addition, XtreamIO is deeply integrated at the product and technology levels, including VMware, VCE integrated systems, VPlex, PowerPath, EMC remote security support, etc.

EMC XtreamIO technical director Ren Yuxiang said: "At present, XtreamIO has more than 100 users, distributed in North America, Europe and the Middle East, most of which are Fortune 500 companies, including world-renowned insurance companies, automobile manufacturers, software vendors, pharmaceutical companies, etc. "

As flash, especially all-flash, becomes popular in the data center, a tiered strategy based on the entire data center will undoubtedly become an element of user consideration. Ren Yuxiang said: "At present, XtreamIO first generation products are not part of EMC tiered storage. In the future, XtreamIO will become part of EMC tiered storage strategy. XtreamIO is an all-flash array, designed in part to solve a specific set of problems where one common thread is that customers can't tell which data is hot and which is cold. A very important premise of tiered storage is that customers need to understand the hot and cold of data. If 90% of the data is hot, it is a completely random read and write environment and scenario. In this case, traditional tiered storage is difficult to solve performance problems. Many customers tell us that their application scenarios do not distinguish between hot and cold data, but they feel that these application scenarios are often the applications with the highest IO requirements, and the future trend is that the proportion of such application scenarios will increase. From three to five years to ten to six months, all data center storage will become flash, that is, all-flash data centers. "

At the same time, Ren Yuxiang also refuted the need for layering within flash memory. "The layering between SLC and MLC was chosen because there was an architectural problem where some writes could not be placed on MLC, and if they were placed on MLC, MLC would be exhausted very quickly," he said. If the architecture is leading, MLC can be used all because MLC is low in price and high in density, so there is no reason not to use it. "

Figure 11: EMC has introduced a version of VCE with integrated XtreamIO all-flash arrays.

Summary: EMC XtreamIO received a lot of attention and controversy when it was launched. Although the first generation product was not perfect, XtreamIO's innovation at the software and architecture level made XtreamIO set a new benchmark in the all-flash array field. There is no doubt that the all-flash array market will become more competitive in the future as EMC officially enters the all-flash array market!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.