Overview of the HP 3PAR Storage Line
Often, the choice of storage is similar to the choice of a cat in a black room, and I really do not want to, after a couple of years of work, it turns out that long-obsolete technologies are hiding behind a beautiful facade, not ready for new trends and not having a "margin of safety" to scale in the future.In our opinion, the 3PAR storage system turned out to be quite successful and having the potential to implement new fashion trends in the data storage market, such as online deduplication, data compression, etc. Whether we are right or not, time will tell. First, a few general words about 3PAR storage systems. The presentation from which the images for this article are taken - just a couple of weeks as published. We mainly touch on the changes and innovations in the 3PAR system in the general system and the storage infrastructure.
What good is the new HP 3PAR system?
- about this under the cat')
Let's start with what any customers start with: “Give me 50TB of disk space! And that's it. ”And what kind of space? how will it work? - the administrator begins to “rake” all these problems after.
To avoid such situations, we recommend that you approach this issue correctly and thoroughly. There are many pitfalls in storage systems, for example:
• Terrabytes against Tibibits.
• The parameter from which juggles many programmers - IOPS (I / O parameter) is a very important parameter, especially for the database.
• Read-write ratios - how does the database work? - will she read more or write more? or is it a video surveillance system that always writes and never reads? what is the size of the block? Is the delay time valid for your application? - etc.
There are a lot of these values. It is very easy to get on the wrong storehouse and then suffer with it for a long time.
Let's start tb vs tib
About these binary units. If the volume is calculated in the wrong units, then with a large number of disks you can make a mistake by a whole order - and these are thousands of IOPS and TB of usable capacity incorrectly calculated in marketing units. Therefore, as always, everything is hidden in the details. And such details, or “pitfalls”, are quite enough for an inexperienced customer to choose for himself a storage system by overpaying substantial money or by receiving a storage system that does not fulfill the tasks assigned to it.
How does Hewlett-Packard propose to solve this problem? For this, HP has an initiative called Convergent Solutions (in the case of storage Converged Storage).
In a nutshell: this initiative lies in the fact that Hewlett-Packard, as a vendor, already offers READY hardware systems for specific tasks. At the same time, the performance of the solution under the stated task is guaranteed by the manufacturer (that is, Hewlett-Packard). HP has a whole portfolio for specific tasks.
Read more about the Convergent Solutions initiative .
Of course, for a convergent solution, you will have to “slightly” overpay, but the customer will immediately have a “headache” when selecting a workable solution, selecting cables, disks, etc.
But back to our main theme - the new model 3PAR.
What is important should be able to modern storage system?
Any customer expects from their storage:
• accelerated I / O operations and full unloading of RAID operations
• fault tolerance with the notorious five nine (99.999%)
• support for performing multiple tasks without the mutual influence of one task on another
• support of any "fashionable chips" - storage levels (Tiering), deduplication (deduplication), data compression, etc.
• support for data manipulation - creating snapshots, clones, online data transfer, etc.
So, in 3PAR, the basic “desires” of the customer are APPARATELY accelerated by a special microcircuit.
We do not need to talk about the advantages of hardware acceleration, it is known to all owners of video cards.
Chip 5th generation (released in July of this year). He does hardware acceleration on the raid.
As you all know, in RAID above level 1 you need to write checksums. If this is entrusted to the central processor of the controller (usually this is IntelXeon, 3PAR is not an exception), then besides its main management tasks, it is necessary to make low-level real-time checksum errors. This leads to "subsidence" performance. In the case of 3PAR, this problem does not exist, because the RAID miscalculation is handled by the 3PAR ASIC, completely unloading the CPU.
The next thing the 3PAR chip does is tracking blocks of zeros in the data stream. Why track zeros? The thing is that the detection of zeros is the core of a fashionable technology called thin volume selection (ThinProvisioning). Its idea is that only useful data is stored on the storage system, and empty disk space (which actually represents zeros) is discarded during the input-output process and not stored on the storage system. 3PAR can detect zeros by a 16kB microscopic block, which is significantly less than that of all competitors, and detection is carried out in real time. What this means for the end user: significant savings in disk space, acceleration of disk formatting in VMWare, automatic release of storage space in the event of data deletion, etc.
The 5th generation chip has learned to detect not only blocks of 16 kB zeros, it can also detect any blocks of 16 kB each. Accordingly, the process can be, as in the archiver, as in the same WinRAR. The controller creates a dictionary of blocks, the first block of 16 kB is called 1, the next 16 kb is called block 2, and so on. D. Further, if the controller encounters a familiar block, then it is not written to disk, but is replaced with a reference to its number, and this is deduplication. Those. if 3PARvidit 16 kb of repetitive data, he finds in his dictionary, then he simply does not write them to disks. The chip can do this in real time, and this is online deduplication. Imagine: there is combat input-output, and 3PAR online deduplicates this information! Nobody can do this yet, only 3PAR.
The task of the 3PAR ASIC is also to connect controllers into a single storage system. As you know, any storage system should have at least 2 controllers. Why? For fault tolerance, plus more, since there are 2, then you can simultaneously load them 2, as a result, the speed of work increases by 2 times. The chip supports combining (already!) 8 controllers into a single storage system. And each controller works as an active-active. Those. we do not accelerate input-output by 2 times, but 8 times accelerated (of course, when 8 controllers are installed).
The next task of 3PAR ASIC is to support heterogeneous load (wide-striping)
The picture on the example of the road is well shown to what this may lead to an unbalanced use of resources. We see a huge queue of cars, although neighboring gateways are walking. Although with the right (intellectual) redistribution of space would be enough for everyone.
How does 3PAR do this intelligent wide-striping?
In contrast to the classical storage system where all data blocks are processed in the order of their arrival (no wide-striping), 3PAR divides each incoming block into the data itself and a control command that says what to do with this data (metadata). Two data streams are formed: control commands (metadata) and the data itself, the control commands (metadata) are processed separately by the central processor of the Xeon controller, and the data itself is processed by 3PARASIC, and Xeon redistributes the data based on the received metadata so that we get the wide striping.
Accordingly, 3PAR at the HARDWARE level is able to balance the load between different tasks, perfectly coping with heterogeneous tasks within the same repository.
Good results with heterogeneous load recognize even competitors (in the face of Oracle) as evidenced by their reports.
The support of the 3PAR ASIC deduplication chip on SSD drives, in addition to saving disk space, also saves money. After all, the cost of a modern solid-state disk for a storage system is comparable to the cost of a car and saving a few dozen gigabytes is already a significant money saving.
Moreover, deduplication itself does not introduce any delay, it occurs online, in real time at the level of the 3PARASIC chip.
Learn more about "fashionable chips", and more specifically about the technology of storage levels (Tiering)
As the study of the average task (for example, a database) shows, in terms of data storage, a very interesting picture emerges (the Pareto principle) - 50% of I / O falls on only 5% of the disk space allocated for the task. This means that by adding only 5% of very fast disks to the storage system (now it’s an SSD), we can get a performance increase of tens or hundreds of times!
The problem is how to calculate these 5% of the data. The Tiering technology successfully implemented in 3PAR helps us in this.
3PAR can divide its disk volume into two or three levels (as practice shows, it is no longer necessary). The first level is usually SSD drives, the second is regular SAS 15K or 10K drives, the third is slow but capacious NL-SAS (SATA) drives.
Accordingly, the first level is the hot data level that provides the fastest I / O. The second level is the level of frequently requested (but not always) data that provides average I / O speed. And finally, the third level - the archive level, which has the largest disk volume, but very modest performance indicators.
Moreover, 3PAR can automatically transfer blocks of “hot” and “cold” data between storage levels. This technology in 3PAR is called Adaptive optimization. Of course, the blocks can be transferred manually, then this is called Dynamic optimization in 3PAR.
Moreover, in 3PAR it is not necessary that each layer be based on a different disk technology. You can, for example, as the first layer select storage on RAID1, the second on RAID5, and the third in RAID6. You can even create levels based on the geometry of the disks: the first level is the inner area of ​​the disks, and the second is the external sectors of the disks.
The true architecture of combining storage system controllers Active / Active
All storage vendors together shout that their storage systems have Active / Active controller architecture, but in practice it turns out that this is an ALUA architecture, which has a low-cost implementation, but with a significant drawback: any storage volume is served by only ONE controller, all other access to this, they do not have (due to the simultaneous recording conflict).
In the case of 3PAR, this restriction was circumvented due to the cunning organization of disk space. In 3PAR, all controllers have access to the same disk volume, and the more controllers in the system, the more productive this disk volume. All controllers are integrated in the full-mesh topology, where the controller from another branch can easily read data from the RAM of another controller - they all represent a single cluster.
What else recently appeared in 3PAR?
Now I have a big fashion for converged storage systems. In this case, convergent means that the storage system provides, besides the classic block I / O via the SAN protocols, also file access via the usual Ethernet. All popular distributed file systems CIFS (Samba), NFS, etc. are supported as network balls. Convergence support is available in all 3PARs with an index with at the end of the model name (7200s, 7400s).
For file systems, all modern "chips" are supported. For example, declared support for SMB 3.0.
Setup and Management
Innovations have also touched the management interface. The new web console is part of the comprehensive OneView management software capable of managing not only storage systems, but also regular servers, blades and chassis, switches and intelligent PDUs through a single console.
The main idea of ​​the new console is maximum ease of management, for example, you can allocate volume to 3PAR using flexible, intuitive policies.
For example, your IT director does not understand what raids are, he only knows that there are different drives in terms of cost: fast, medium and slow drives, and he, or someone from the management, needs a visual scheme - what to what You can create different policies for yourself, for example, you create policies like Gold (expensive and faster disks), Silver (slow and cheap disks), etc. You can create such a variety of policies the right amount and then “drive” your storage system with an economy based on them. This ideology is very convenient for cloud solutions.
About the reliability of 3PAR
Hewlett-Packard is so confident in the reliability of 3PAR storage systems that it has launched an initiative called “HP 3PAR Get 6-Nines Guarantee Program”. In a nutshell, this initiative describes the fact that HP is ready to go for penalties in the event of a breach of reliability of 99.9999% (a mere 30 seconds per year) for 4-controller 3PAR storage systems.
For more information about this marketing initiative, click here.
What makes such confidence so?
It is based on a number of architectural features of the hardware component 3PAR. The first is the lack of a dedicated disk for hot rebuild RAID, the so-called Hot-sparedrive.
For a classic Hot-spare storage system, the disk is usually a bottleneck when recovering from a failed RAID. All blocks that were lost during a failure are restored to one disk! This is still valid when recovering a single RAID, but if you need to restore two, three or more failed RAID? Experienced administrators have reported cases that when a SATA disk fails to a few terabytes in size, during recovery (usually dozens of hours), the second SATA disk usually fails due to the increased load. That is why all vendors like one recommend only RAID 6 on slow NL-SAS (SATA) drives.
In the case of 3PAR, there is no such problem, since 3PAR does not create RAID on the disks, it creates RAID on the blocks (in modern 3PAR this 1GB block), respectively, repairing a failed RAID can be done to any free block on ANY disk that has space at least 1 GB.
The next fault-tolerant feature is the fault-tolerant cache (in terms of 3PAR Resilient Caching). It is not necessary to explain that the cache on any system significantly speeds up the input-output processes. 3PAR is no exception, there are several tens or hundreds of gigabytes of cache memory on each storage system controller. So, in the case of the classic dual-controller storage system, the failure of one of the controllers leads to automatic shutdown of the memory cache, due to the emergence of a single point of failure (there is no place to mirror the cache on the record, because the partner controller failed). This leads to significant performance degradation, which naturally affects business processes.
In the case of 3PAR, this problem is absent, because 3PAR can have more than two controllers, respectively, there is no single point of failure, and the cache from the failed controller is simply redistributed between the surviving storage system controllers
The third fault-tolerant feature is the high-level 3PAR PeerPersistence software, which is based on the VMware operating system and allows you to get a geo-distributed fault-tolerant cluster out of the box without the “torment” with the VMwareSRM setting.
This is the most purchased software option that goes together with 3PAR, which allows you to get out of the box a disaster-proof solution based on two distance-separated data centers. About the seriousness and maturity of this decision says at least long-term cooperation of 3PAR developers with VMware. For example, VMware VVOL was created in the image of the internal data storage architecture in 3PAR.
Consider the entire 3PAR storage family available on the market.
The first thing I want to say is that 3PAR is a mid-range and high-end (Hi-End) storage system. Mid-level models are represented by the 3PAR 7000 family, and older models are 10,000 and 20,000. The difference between the models is only in hardware, the operating system and the functionality is the same for all models. Accordingly, having mastered the work with the youngest system 3PAR 7200, you can safely manage the most senior “monster” 3PAR 20000 scalable to more than 1000 disks.
Each family has models that end in 50 (For example, 3PAR 7450). These are special models optimized to work exclusively with SSDs. As it turned out, the internal architecture of 3PAR is very suitable for working with this type of disks.
The difference between these models and index-free counterparts is only in the power of the installed Xeon and the cache size.
Also, SSD drives can be used as an extension of the cache, increasing the amount of embedded cache to hundreds of gigabytes. This functionality in terms of 3PAR is called AdaptiveFlashCache, it is free on all models and is activated by a pair of commands. What is most interesting, under the cache, you can not allocate the entire volume of the SSD drive, and use the rest for normal RAID.
Naturally, the speed of work with the activation of Adaptive Flash Cache increases significantly, which is confirmed by testing.
3PAR and data backup
If you are thinking about fast backups (backup), then the connection 3PAR with HP StoreOnce backup system turns this bundle into an ultra-productive tape library that allows you to backup hundreds of gigabytes in minutes. At the same time, 3PAR acts as a repository, and StoreOnce emulates a tape device. This emulation is necessary because All the usual backup programs are used to working with tape drives, and there is little point in “retraining” them.
Finally, if you decide to switch from your old storage system to 3PAR, Hewlett-Packard will help you with the built-in 3PAR utilities. This functionality is called 3PAR Online Import and allows FOR MOVE (i.e. without stopping the business process) to migrate disk volumes from the old storage system to 3PAR. EMC, Hitachi storage systems and older HP EVA systems are currently supported.
Well, if so briefly, HP did, as we consider very worthy storage system. If you fully describe in detail all the technologies that work in 3PAR, and in detail about what 3PAR can do, there are not enough 4 such articles.
UPD on holivaru in the comments:
For those who doubt the efficiency
UPD2:
The presentation, from which the most recent images are taken to the review, can be confusing to the fact that the incorrect footer from PowerPoint got into the screenshot. The functionality described in the article (deduplication, for example) appeared with the release of the 5th generation ASIC (beginning of this month).
Distribution of HP decisions in Ukraine , Georgia , Tajikistan , CIS countries .
Training courses on HP technologies in Kiev (TC MUK)
MUK-Service - all types of IT repair: warranty, non-warranty repair, sale of spare parts, contract service
Source: https://habr.com/ru/post/263469/
All Articles