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

What this article shares to you is about how to interpret the details of Dell EMC PowerMax. The editor thinks it is very practical, so I share it with you to learn. I hope you can get something after reading this article.

As an official account called "high-end storage knowledge", it will be incompetent if you don't interpret Dell EMC's latest release of PowerMax. Although there have been a lot of posts online, many netizens still want to see what Brother Watermelon thinks and how to evaluate this monster instead of high-end storage VMAX.

Dell EMC PowerMax

I actually had a little expectation for EMC to launch this new product. Because when DSSD was delisted, EMC indicated that its technology might be used in the next generation of high-end storage, so it should be sooner or later for high-end storage to support NVMe.

I just didn't expect that EMC had taken such a big step. Now you go to EMC's product page and you can't find VMAX, and even VMAX-AF is gone.

Of course, you can still see it if you use a search. It shows that VMAX and VMAX-AF are still on sale, but not the main push, the future of high-end storage only belongs to PowerMax.

In fact, according to the data of Gartner Q4 in 2017, I counted that EMC's VMAX-AF already accounts for 3x4 of high-end storage.

That is, the high-end storage form is now basically all-flash form.

To tell you the truth, looking at the trend, you have to admire Gartner. In a report by IT Market Clock for Storage in 2016, Gartner predicted that high-end storage would retire soon.

That is, high-end storage is 2-5 years away from the Replacement stage, at this stage, users should not buy new high-end storage. Two years after 16 years, isn't it 2018? EMC's pace seems to have stepped on a bit.

This time the PowerMax uses the NVMe architecture, which is also predicted by Gartner. Gartner thinks the NVMe back-end will be adopted in 2018 and the front-end NVMe-oF in 2019. The mainstream NVMe-oF protocols in the future may be FC and RoCE. The same seems to be true of the road signs in PowerMax.

In fact, Huawei launched NVMe products in the first half of 2017 and Pure Storage in the second half of the year. PowerMax will be launched in 2018. In terms of stepping point, I only take EMC,O (∩ _ ∩) O ha!

OooO ↘┏━┓ ↙ Oooo

(step on) →┃ you ┃ ← (die)

\ (→┃√┃ ←) /

\ _) ↗┗━┛ ↖ (_ /

All right, let's get back to business, let's interpret this PowerMax.

First of all, there are only three keywords for PowerMax: speed, intelligence and efficiency.

Compared with VMAX-AF:

1. SAS becomes NVMe, and the performance is greatly improved, so it is faster.

2. With the introduction of AI/ML, it has a certain ability of automatic operation and maintenance, so it is more intelligent.

3. In addition to compression, there is more support for re-deletion, so it is more efficient

Specifically, at Dell Tech World 2018, EMC released two PowerMax,2000 and 8000. The name Power probably comes from the Dell server PowerEdge, there is no way, because now the storage leaders are basically Dell people.

I compared, in addition to fast performance, the software features are basically the same as the original VMAX-AF, but the lack of a CloudArray cloud gateway function, it is estimated that it will be made up later.

CPU is still v4, without the latest v5. The port is still 16G FC, and it does not support high-speed interfaces such as 32G FC,25G. This road sign looks like 2019.

In terms of indicators, we can also see whether the maximum supported effective capacity is the same or the most 4PBe. According to the definition of EMC, this effective capacity is the capacity that contains thin, re-delete compression and snapshots. In other words, its physical bare capacity is actually not too large.

If you look at my previous analysis, you can see that this is a restriction inherited by VMAX. Because the maximum effective capacity supported by VMAX is related to the size of memory, now the largest amount of memory in PowerMax is 16TB, just like VMAX, so the maximum effective capacity supported by 4PB can only be 4PB.

Next, I will talk about these three keywords respectively.

First, fast

EMC claims that PowerMax is the fastest storage in the world, to be honest, if in China, we can sue EMC for violating the advertising law, O (∩ _ ∩) O ha!

However, from the index point of view, it is really fast, such as 10 million IOPS, although all are measured under the memory hit, there is no reference significance.

However, the delay still means something. Compared with VMAX-AF, the latency is reduced by 25%, and if SCM is used, it will be reduced by half.

The reasons for being quick are NVMe and SCM.

Second, intelligence

Now you release a product, not to mention AI is shameful. PowerMax is no exception. That is, the storage will be on autopilot in the future.

At the press conference, EMC piled up a pile of data, in a word, PowerMax can handle so much data, otherwise how could it be called big data,O (∩ _ ∩) O ha!

However, after blowing so much, the real scene is mainly in the intelligent layout of the data. That is, analyze the application, and then you know whether the data is more appropriate on SCM or Flash. However, the feature of EMC will not be available until 2019, and it is only advertised in advance, because SCM is not supported yet.

However, this idea was not invented by EMC, but by Moshe Yanai, the father of high-end storage. He designed the third generation of high-end storage Infinibox, using AI to do Cache, known as "neural caching". In my article, AI has three landing applications in enterprise storage, do you know? It was introduced in.

However, at present, PowerMax's cache algorithm does not have AI function. I read EMC's data that PowerMax's Cache hit rate is only 50%, while other people's Infinidat advertises that the Cache hit rate of their own neural cache algorithm can exceed 98%. Maybe the PowerMax AI version will be launched next year to improve this indicator.

Another place to use AI is in the compression section. If using AI to analyze some data is hot data, there is no need for compression, because repeated compression and decompression is still a waste of a lot of resources. This also makes sense. It was not mentioned in this material. I only knew it when I was interviewed by the media. Pity my broken English. I hope I heard it right. However, this also needs to wait for the AI version to come out.

In a word, intelligence is a signpost, and PowerMax is not smarter than VMAX-AF yet.

Third, high efficiency

As we all know, VMAX-AF has a compression function, but PowerMax has a re-delete function, so it claims to be able to achieve the 5:1 data reduction ratio.

Efficiency is also reflected in the saving of space.

And the reduction of power consumption.

However, Pure Storage expressed disapproval in this area. Pure / X uses CFM, self-made NVMe board, higher density, and Pure data reduction thinks it does better than PowerMax, so it saves more space. However, it is a bit funny to compare other people's cabinets, because that cabinet is still very empty and can be used for other equipment.

After talking about the three key words, I would like to see some implementation details, especially those that are different from VMAX-AF.

A 、 Smart RAID

First of all, let's talk about the dual-port NVMe that EMC always emphasizes, which is nothing special. For the sake of reliability and performance, NVMe SSD had better be dual-port, and the NVMe SSD used by Huawei's Dorado v3 is also dual-port. Generally, consumer-grade NVMe is single-port.

Because it is dual-port, each controller (Director) can access one port. Therefore, EMC also wraps the term Smart RAID, and because of the parallelism of the NMVe protocol, each controller can access all SSD at the same time, whether or not you are in a RAID group.

In fact, the SAS disk is also dual-port, but because the SAS/SCSI protocol does not have parallel capability (a LUN only supports one queue), it is impossible to share the RAID group of Active/Active.

2. The connection of NVMe disk frame

EMC uses PCIe on the back end of PowerMax, instead of using RoCE to connect as many Startup companies do.

Since none of the original storage has a DAE that supports PCIe, this is brand new hardware. The EMC is also designed as a 24-disk, 2U-high DAE.

In the case of a PowerMax 2000, an engine (2 controller) can connect two DAE. 2000 supports up to 4 controls, so the connection method is as follows:

But if it is a high-end PowerMax 8000, the connection is strange. 2 engines (4 controllers) support up to 3 DAE, that is, one DAE in the middle is shared by both engines.

After taking a look, I thought that EMC also made a HDS and Huawei boast of 4 bad 3 designs, that is, the data of the middle DAE is very safe, because any engine (2 control) failure, the other engine is also connected to the circuit, that is, it can be accessed?

Actually, that's not the case. If you take a closer look at the middle DAE, it is actually divided into two parts. The first 14 positions are controlled by engine 1, and the last 10 positions (numbered 15-24) are controlled by engine 2. That is, if engine 1 fails, the disk managed by engine 1 cannot be accessed by engine 2.

We can see that the middle four bits (No.13-16) of the middle DAE are used specifically for hot spares, that is to say, the RAID mechanism of PowerMAX is the same as that of VMAX, and special hot spares are still needed.

Why the middle DAE is not divided equally, but the first 14 disks are given to engine 1, and the last 10 disks are given to engine 2. It is estimated that it is related to the design of the RAID group. EMC generally uses 8 disks and a RAID group, 7-1 or 8-2. However, the maximum effective capacity supported by PowerMax is determined by the size of memory, so the bottleneck is not on the number of disks. I used PowerMax Studio to do a maximum configuration, you see the lower part of the DAE disk is how to insert (the upper part is the same), very irregular, and there are a lot of empty positions.

C, re-delete compression

EMC's PowerMax has been re-deleted this time, but the compression function is also different from the original VMAX-AF. However, both are achieved through special hardware acceleration cards.

Look at the engine, the NVMe back-end card and data reduction card two types of cards are new, the other cards are the original VMAX-AF. The Infiniband connection is still used between engines.

This module is not only responsible for the re-deletion and compression of written data, but also responsible for data compression during SRDF replication.

This is a comparison of re-deletion compression features.

We see that compression adds an EDC, that is, the extended data compression function. This function is to see if the less commonly used data can be further compressed when the system is not busy, so that a better compression ratio can be obtained. I actually learned this idea from Pure Storage.

Also, the compression algorithm is different. PowerMax uses DEFLATE algorithm, while VMAX-AF uses LZS algorithm. Specific differences between Baidu bar, the following is a list of the development of lossless compression algorithm. I'm not very good at math and I don't know much about the differences.

However, the compression of PowerMax is not good. It is estimated that in order to ensure the compatibility of SRDF, when data is copied remotely, it needs to be decompressed first, and then compressed by SRDF algorithm. After receiving it, there is a process of decompression and re-compression, which is a waste of resources.

After talking about compression, let's take a look at re-deletion, this is a new feature. From the process provided by EMC, we don't see the same step as hash ID and then compare the data.

In other words, PowerMax, like XtremIO, uses a strong hash algorithm (the document says it uses SHA-2 algorithm) and does not compare the way data is processed. This is different from the thinking of NetApp and Huawei. Theoretically, there is a risk of hash conflict.

D, SLA service model

This is the same as VMAX, which uses SLA for hierarchy. However, VMAX adjusts the service level by adjusting the flash percentage of LUN, while PowerMax is all NVMe, so it can only be slowed down by controlling the Host O Limit of the front-end port.

Look, it's all NVMe SSD, but it also provides bronze service with 7.2ms delay. These are mainly for ISP, because some users want cheap services and need to bring down the experience, otherwise they will not be able to sell expensive services.

E, second-level NDU

PowerMax inherits the reliability of VMAX, even the power of NDU in seconds.

The principle is that when upgrading, the code is loaded into EEPROM in advance, and then all the controllers are hot loaded at the same time, the overall system only needs about 10 seconds to complete the upgrade, and the business is not interrupted.

The ownership of F and LUN

We know that VMAX has no LUN attribution, that is, LUN can be read and written by multiple controllers at the same time, and LUN does not need to be forwarded. However, LUN does not belong, and one of the following logical blocks still needs to belong. HPE 3PAR is CK, and VMAX is TDAT.

SmartRAID's description mentions that TDAT belongs.

The advantage of this design is that the host can only see the LUN, and the lock mechanism is not needed if the TDAT,TDAT does not belong, so the system performance is better, and the host can balance the multi-path load of the same LUN, the controller is not aware of the host, and the business impact is small.

Well, at this point, I have finished interpreting all the technical points I can think of. Finally, let me ask you a question: according to the definition of IDC, is PowerMax native or improved?

Indeed, PowerMax has a lot of new things, but we see that the software is basically the same as VMAX, especially the implementation of RAID and SRDF compression, and we all see a lot of compromises. In other words, there is still a historical burden, and it is impossible to be as light as a newly designed architecture. However, the improved model also has the advantage of rich functions and high maturity. Of course, the downside is that there is a compromise, performance can not be played to the extreme.

The above is how to interpret the details of Dell EMC PowerMax. The editor believes that there are some knowledge points that we may see or use in our daily work. I hope you can learn more from this article. For more details, please follow the industry information channel.

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.