ALLONE Cloud Disk Drive 101 RAMDisk Review (32GB) – 500K IOPS of DDR3 Storage
Source: | Author:pmo44feef | Publish time: 2014-07-17 | 1859 Views | Share:
It’s not every day that we see an SSD that makes us stop and think. After our trip to Computex, we were able to take delivery of something that is fairly unique in terms of enterprise storage. The ALLONE Cloud Disk Drive 101 RAMDisk is not your normal SSD, in fact, it really isn’t an SSD at all. Technically speaking, it is a battery backed, PCIe-based RAM disk. RAM disks are nothing new. They have been around as long as we can remember.
It’s not every day that we see an SSD that makes us stop and think.  After our trip to Computex, we were able to take delivery of something that is fairly unique in terms of enterprise storage.  The ALLONE Cloud Disk Drive 101 RAMDisk is not your normal SSD, in fact, it really isn’t an SSD at all. Technically speaking, it is a battery backed, PCIe-based RAM disk.  RAM disks are nothing new. They have been around as long as we can remember.



The intent is to, through software, take a portion of system memory and present it to the OS as a storage device.  By doing this, you create a tiered storage solution where you can cache frequently used data and access it at DRAM speed.  They were more popular at a time where storage was painfully slow when compared to CPUs and RAM.  With SSDs, specifically PCIe SSDs, storage speed has increased to a point where storage is less of a bottleneck, but it is still nowhere near the speed of system memory.



The issue with traditional RAM disks is that they are volatile.  When power is removed, the data on them disappears.  They also take up valuable system memory and bandwidth.  One solution is to take the DRAM off of the memory bus and put it on a PCIe add-in card, which exactly is what ALLONE has done.
Now, this is not a new idea.  There have been multiple companies over the years that have tried, unsuccessfully, to push this type of product to market.  For the most part, those products still used a SATA interface, which greatly limited the effectiveness of the solution. The DDRDrive X1, ACard’s ANS-9010 and Gigabyte’s i-RAM are all examples of solutions that failed to gain acceptance.  ALLONE hopes their Cloud Disk Drive 101 offers performance that can’t be overlooked.

SPECIFICATION
The Cloud Disk Drive 101 comes in two flavors.  Today we are reviewing the Storage Turbo, which comes equipped with 32GB of capacity in the form of 4 Kingston DDR3 modules.  Those modules are backed up with a 3000mA battery backup.  In case of power loss, the DRAM is backed by 6 Kingston microSD cards, for a total of 48GB.  The Cache Pro solution is similar, but lacks the SD backup capabilities and has lower overall performance.



Random performance is impressive at nearly half a million IOPS, but that is at 512 bytes.  We are used to random performance rated at 4KB workloads, so we will have to see how the Cloud Disk Drive 101 handles larger workloads.  The sequential numbers are a bit low, but considering the PCIe x4 Gen 1 interface, which has a real-world bandwidth of 800-900MB/s, it is not surprising.
There are a few advantages that this type of architecture has that typical SSD don’t.  The first, and by far the most important, is write endurance.  Even the very best SSDs on the market have a finite life based on bytes written.  The Cloud Disk Drive 101 has no such limitation.  Theoretically, DRAM has infinite write endurance.  As a caching solution, you will never have to worry about the type of workload.

Secondly, the Cloud Disk Drive has a very low power consumption.  Typically, products that produce this level of performance are PCIe devices that can consume upwards of 25W.  At slightly over 7W of active power draw, the IOPS/W ratio is excellent.
But, all of this comes at a price.  The Storage Turbo will retail for $15,000 while the Cache Pro is $6,000.  Those prices will be hard to swallow for consumers.  What remains to be seen is whether enterprise customers can see value in this solution.

INTERNAL COMPONENTS


As we eluded to earlier, the ALLONE Cloud Disk Drive 101 is made up of 4 major components: DRAM, Controller, Battery and SD Card backup.  These components are all located on a full height PCIe x4 Gen1 add-in card.


For the DRAM, ALLONE has chose Kingston’s HyperX PnP.  This SO-DIMM DDR3-1600 module is of the 1.35V variety.  In total, there are 4 SO-DIMM slots, giving us a total capacity of 32GB.

Next up is the RamCore M5338e controller.  This controller manages the DRAM and SD backup operations.



The battery backup solution is a large 3.7V 3000mAh external battery that is held in place by a plastic bracket.  The battery serves two purposes.  The first is to maintain the power for the system when there are brief power failures.



The second is to hold that power on long enough to back up to the microSD cards.  There is an LED indicator to show when batter power is being applied.


Finally, we have the 6 microSD cards, which are also Kingston.  These 8GB modules are all class 10 and are in a RAID 5 configuration, so the loss of a single card does not cause data loss.



Each microSD slot also has an LED to indicate whether the card has been removed.  It is interesting that ALLONE chose microSD cards for their backup and not more reliable on-board NAND.  ALLONE made it clear to us that the microSD cards are for emergency situations only, so they should not see very much usage.

TEST SETUP AND METHODOLOGY
In testing the ALLONE Cloud Disk Drive 101, along with all enterprise drives, we focus on long term stability. In doing so, we stress products not only to their maximum rates, but also with workloads suited to enterprise environments.
We use many off-the-shelf tests to determine performance, but we also have specialized tests to explore specific behaviors we encounter. With enterprise drives, you will see that we do not focus on many consumer level use-cases.


Our hope is that we present tangible results that provide relevant information to the buying public.


LATENCY
To specifically measure latency, we use a series of 512b, 4K, and 8K measurements. At each block size, latency is measured for 100% read, 65% read/35% write, and 100% write/0% read mixes.



For average latency, the Cloud Disk Drive 101 gave us some really interesting results, especially the 4KB workload.  As you can see, the results for read, write and mixed are almost identical.  That is definitely not normal for an SSD, but for a DRAM based product with symmetrical  r/w performance, it looks about right.  The CDD 101 also excels at small block transfers.  As you can see, the 512B performance is one of the highest we have ever seen.

Maximum latency was also excellent.  Typically, we see maximums in the 5-20ms range.  For all of our tests, we never topped 1ms, which is very impressive.

ADVANCED WRITE TESTING
As we talked about in our Micron P400m SSD Review, SSDs have different performance states. Since the ALLONE Cloud Disk Drive 101 is an enterprise storage device, we will focus on steady state performance. With the following tests, we stressed the drive using random 4KB write workloads across the entire span for at least 24 hours. This is more than enough to achieve steady state. The following graph is showing the latency and IOPS across an 11 hour span.



Here is the deal with the ALLONE Cloud Disk Drive 101; it behaves so unlike an SSD that many of our long-term tests are unnecessary.  For a NAND-based SSD, we spend countless hours coaxing them into steady-state.  Because the CDD 101 does not have NAND and doesn’t have to worry about P/E cycles, there aren’t different performance states, just steady-state.  That is a major benefit of this product is that whether you are 1 second or several days into a run, the performance is identical.




Since we normally run our SSDs with a 4KB workload, the ALLONE does not show its 470,000 IOPS that it can do at 512B.  But, what we did see was a very respectable 118,000 IOPS.  The performance was also very consistent, with just a few data points in the 120,000 IOPS region.
This was probably one of the easiest storage products to test that we have even had in our labs.  It would be reassuring, as a customer, to know that your performance is guaranteed no matter how long your run your test.

MEASURING PERFORMANCE
Normally we give a boilerplate explanation of how we secure erase the drive and make sure it is in steady-state before we begin testing.  Since the ALLONE drive is always in steady state, we just moved right into our performance testing.




For random 4KB write operations, the ALLONE did a better job at staying with the pack.  Once the queue depths were sufficiently high, it easily outpaced the Intel and Micron offerings.

Random 4KB reads were almost the complete opposite.  At low queue depths, it was free and clear of the other drives, but stalled as queue depths rose.  Once again, while the competition was putting up 300-750K IOPS, the ALLONE was stuck at 130K.




As we mentioned earlier, sequential performance is not the strong suit for the Cloud Disk Drive 101.   While it was able to hang close with the Micron P420m for sequential writes, it was no match for the competition for the remaining tests.  In fact, it was only 10-20% than most enterprise SATA SSDs. We are so used to PCIe-based storage products hitting 1 or 2 or 3GB/s in our sequential tests, that we were a little disappointed.  Fortuantely for ALLONE, sequential operations are not the intended workloads for the Cloud Disk Drive 101.

At this point, we aren’t quite sure what to think about this drive.  While it has some unique characteristics, it’s losing certain tasks by a very wide margin.

SNIA IOPS TESTING

The Storage Networking Industry Association has an entire industry accepted performance test specification for solid state storage devices. Some of the tests are complicated to perform, but they allow us to look at some important performance metrics in a standard, objective way.



SNIA’s Performance Test Specification (PTS) includes IOPS testing, but it is much more comprehensive than just running 4KB writes with IOMeter. SNIA testing is more like a marathon than a sprint. In total, there are 25 rounds of tests, each lasting 56 minutes. Each round consists of 8 different block sizes (512 bytes through 1MB) and 7 different access patterns (100% reads to 100% writes). After 25 rounds are finished (just a bit longer than 23 hours), we record the average performance of 4 rounds after we enter steady state.
Preconditioning: 3x capacity fill with 128K sequential writes
Each round is composed of .5K, 4K, 8K, 16K, 32K, 64K, 128K, and 1MB accesses
Each access size is run at 100%, 95%, 65%, 50%, 35%, 5%, and 0% Read/Write Mixes, each for one minute.
The test is composed of 25 rounds (one round takes 56 minutes, 25 rounds = 1,400 minutes)



Once again, we have to remind ourselves that we are not dealing with NAND storage.  There are two things that stand out that make these graphs unique.  The first is that, other than 512B, there is almost not difference between read, write and mixed workloads.  Take another look at the bar chart, they are almost completely identical.
    
The second thing that jumped out as we reviewed our results is that the performance is the same, regardless of previous operations.  We normally run our SNIA tests because it helps show how an SSD handles quick transitions between different workloads.  Because of this, you sometimes see results that are lower than if you ran that individual test for long periods of time.  With the ALLONE, much like rotating HDDs, the performance you get is completely independent from any previous operations.  This is a trait that most SSDs would kill for.

SERVER PROFILES

While synthetic workloads do a great job of testing the underlying technology and reporting easy to understand results, they aren’t always indicative of how the drive will be used by the end user. Workloads that simulate enterprise environments try to bridge that gap without being overly complex.



The database profile is 8K transfers, and 67% percent of operations are reads.

At low queue depths, the ALLONE does a terrific job running our database profile.  Unfortunately, the superior high queue depth read performance allows the competition to pull ahead at the end.



The fileserver profile is based on an 80% read/20% write mix. Its made up of blocksizes from 512 to 64K, each making up a different percentage of the access pattern. The pattern is: 512 bytes=10%, 1k=5%,2k=5%, 4k=60%, 8k=2%, 16k=4%, 32k=4%, 64k=10%.

Not much changed in our fileserver testing.  The ALLONE came out strong in the low queue depths, but didn’t have the read performance to maintain it.



The webserver profile is similar to the fileserver profile, but has some additional 128K and 512K accesses thrown in for good measure. Additionally, the profile is 100% read.

This is probably the worse case scenario for the ALLONE, lots of random and sequential reads.  It was able to keep a lead for the first few queue depths, but eventually got left in the dust.

REPORT ANALYSIS AND FINAL THOUGHTS

After all of the testing, we came away with a couple conclusions that may seem obvious, but we had to do the testing to prove it.  The first is that the ALLONE Cloud Disk Drive 101 is not an SSD.  Other than the fact that you can write to and read from it, it shares very few similarities.  In some ways, that’s a good things.



We were impressed by the symmetric performance and how previous operation had zero affect on the current workload.  It seriously drives enterprise SSD reviewers up a wall trying to get SSDs to enter and stay in steady-state.  With the ALLONE, that isn’t a factor in any way.

The real issue is that the Cloud Disk Drive 101 is a niche product.  At 32GB, it is limited in size, which will limit its applications.  It performs well in certain areas, but gets thrashed in other areas by PCIe SSDs.  ALLONE has put all of its efforts in touting the drive’s 512B performance, but we aren’t sure how many applications will take full advantage of that.



ALLONE makes it very clear that they are going up against Fusion-io.  When they aren’t calling them out by name in their comparison documents, they are referring to them as Tier-F.  This also helps explain the pricing.  At $15K, the price is going to be a hard pill to swallow for most, but when compared to what Fusion-io charges, that isn’t exactly out of the question.  We have to believe that this is a product that was better suited for 3 years ago.  With NVMe hitting servers as we speak, the gap in performance is closing fast.  In fact, the Intel SSD DC P3700, especially the 1.6TB model, would have beat the ALLONE in almost every way at a third of the price.

After talking with ALLONE, we got a better understanding to where this product belongs.  They want to augment their customer’s tiered storage solution.  They want their product to get in between typical PCIe SSDs and DRAM.  They want to accelerate small block transfers using as little power as possible.   We can understand that.  Their challenge may be convincing customers that their storage rack needs another level of storage.

It’s not all doom and gloom.  If ALLONE can make a few evolutionary changes, such as a PCIe x4 Gen3 interface, more DIMM slots and DDR4 support, that can move that curve back in their favor.  As it stands, this product is for people that need unlimited write endurance for 512B workloads and are willing to pay for it.
热点文章