SSD is not for everyone, or how I talked to Fusion-io ioDrive Duo
We have become accustomed to the fact that everyone is talking about the advantages of SSD (just remembering to intimidate them with their disadvantages in terms of fault tolerance), and we are used to the fact that SSDs are “quick but expensive”. So, the other day just one of these came into my hands - expensive SSD models, in particular, Fusion-io ioDrive Duo. Of course, this is already quite an old model, which appeared on the market in the distant 2009th year, but it can still be found on sale. Just why is this thing so good, and why were there so many enthusiastic screams in the IT press about the fastest SSD (which of course it was 3 years ago)? Let's take a look at it today.
The first minus was “uncovered” in the first minutes after installation - this disk is not detected by the motherboard at all, that is, it cannot be bootable, and the system cannot be installed on it, but can only be used as additional storage for data. For example, the same PCI-E SSD from OCZ is quietly detected by the motherboard and can be used as a boot.
Well, we will look further ... Fortunately, the second minus did not take long to wait: in addition to the driver, which we managed to find on the manufacturer’s website only after a long registration form, we needed firmware compatible with the version of this particular driver, as well as a separate utility Dell IO Manager (why Dell for me personally remains a mystery, since IO Data supplies its models for both Dell and HP and our copy was exactly HP) for managing and maintaining this drive (by the way, with rather poor functionality). So, spent 1.5 hours of time, but so far on their own to make this thing work and see the disk failed. (and some western colleagues could not cope with this task on their own and applied to the manufacturer’s support).
And here is our hero, in all its glory, so to speak.
')
But all these shortcomings could be forgiven him if he manifests himself in the work, because the manufacturer promises us amazing speed characteristics:
As can be seen from the label, in our model 1.28 TB, the MLC memory (made by Samsung) is used, but for some reason the manufacturer has kept silent on which controller is installed in it. Although the autopsy showed that it contains two FPGAs (programmable, logical integrated circuit) Virtex-5, manufactured by Xilinx .
Bearing in mind that this drive contains 2 controllers and is accordingly defined as 2 different disks, I really wanted to find in it the possibility of creating a RAID array of 1 and 0 levels, but ... And then we were severely disappointed - software RAID only .
Let's get down to the most interesting part - the tests, and check what he is capable of, what vertices of reading and writing will he submit to? Due to the fact that it is impossible to create a “hardware” raid array from it, and a soft raid is not perceived by the testing programs, each of the “sections” will be tested separately.
Testing was conducted on a Dell PowerEdge R710 server .
Dell PE R710 Server Configuration:
Do not be surprised that the volume of each partition is 320GB, I set the maximum performance (Write Write) in the disk settings.
Fusion-io delivers its own disks for both Dell and HP , and each company has its own disk manager. In the screenshot above, this is the Dell IO Manager , and the HP manager is the web interface — the HP IO Accelerator Management Server .
On the manufacturer's website, along with the latest drivers, it is Dell IO Manager that lies, and HP IO Accelerator Management Server can be found on the HP website, respectively. It is worth considering that at the time of testing the driver on the HP website was version 3.1.1, which was listed on the official website as Private Version (I understand something like the Beta version), and Current Stable was 2.3.10
Well, as a starting point for testing, I will give the performance of our RAID10 on 6xSAS 146GB 15k:
Well, now the performance of two sections of our experimental ioDrive Duo
We were extremely surprised at such a low disk performance. Rolling back and reinstalling the drivers yielded no results! What does the disk showed the same performance in all modes of Maximum Write Performance and Maximum Capacity. We already had time to think that there might be a problem with the PCI-E port on our Dell server, so we decided to replace the test platform. Important factors for replacing steel: HP sticker on the disk itself, as well as more recent drivers on the HP website, which are not available to the mere mortals on the developer's website. Our new test platform has become familiar with comparative tests - HP DL380 G7 , in the following configuration:
To begin with, the same HD Tune test was launched to check if there is any difference ...
And we saw her!
The difference is almost 2 times! That's what drivers can do. This is how it looks in the diagram:
Please note that the total disk capacity is 512GB, that is, in the Maximum Write Performance configuration, it “eats away” not only 50%, but only 20%, which is also very good. Of course, the failures on one of the sections do not look very good, but there's nothing to be done about this - the lack of support for TRIM makes itself felt. Now we are convinced that everything is in order with the disk, and we can proceed directly to testing, dropping unnecessary conversations.
AIDA64 Disk Benchmark
Crystaldiskmark
AS SSD
Of course, up to the speeds declared by the manufacturer - practically, as before the moon, but we have long been accustomed to the fact that the marketing figures still quite strongly disagree with the realities.
In the near future, we plan to conduct comparative testing of several SSD drives, including both conventional SATA and RAID, and PCI-E from other market players, and we will compare all of this with arrays from conventional SAS 10 / 15k disks. Moreover, many horror stories about the fact that SSD drives are unreliable, have long been transferred to the myths section, and they work no less than ordinary HDDs. Given the warranty period, you can safely rely on these products, but you should not forget about backups anyway!
By KorP
The first minus was “uncovered” in the first minutes after installation - this disk is not detected by the motherboard at all, that is, it cannot be bootable, and the system cannot be installed on it, but can only be used as additional storage for data. For example, the same PCI-E SSD from OCZ is quietly detected by the motherboard and can be used as a boot.
Well, we will look further ... Fortunately, the second minus did not take long to wait: in addition to the driver, which we managed to find on the manufacturer’s website only after a long registration form, we needed firmware compatible with the version of this particular driver, as well as a separate utility Dell IO Manager (why Dell for me personally remains a mystery, since IO Data supplies its models for both Dell and HP and our copy was exactly HP) for managing and maintaining this drive (by the way, with rather poor functionality). So, spent 1.5 hours of time, but so far on their own to make this thing work and see the disk failed. (and some western colleagues could not cope with this task on their own and applied to the manufacturer’s support).
And here is our hero, in all its glory, so to speak.
')
But all these shortcomings could be forgiven him if he manifests himself in the work, because the manufacturer promises us amazing speed characteristics:
As can be seen from the label, in our model 1.28 TB, the MLC memory (made by Samsung) is used, but for some reason the manufacturer has kept silent on which controller is installed in it. Although the autopsy showed that it contains two FPGAs (programmable, logical integrated circuit) Virtex-5, manufactured by Xilinx .
Bearing in mind that this drive contains 2 controllers and is accordingly defined as 2 different disks, I really wanted to find in it the possibility of creating a RAID array of 1 and 0 levels, but ... And then we were severely disappointed - software RAID only .
Let's get down to the most interesting part - the tests, and check what he is capable of, what vertices of reading and writing will he submit to? Due to the fact that it is impossible to create a “hardware” raid array from it, and a soft raid is not perceived by the testing programs, each of the “sections” will be tested separately.
Testing was conducted on a Dell PowerEdge R710 server .
Dell PE R710 Server Configuration:
- Processor - Xeon E5607
- Memory - 16GB
- Hard drives 6xSAS 146GB 15k RAID10
Do not be surprised that the volume of each partition is 320GB, I set the maximum performance (Write Write) in the disk settings.
Fusion-io delivers its own disks for both Dell and HP , and each company has its own disk manager. In the screenshot above, this is the Dell IO Manager , and the HP manager is the web interface — the HP IO Accelerator Management Server .
On the manufacturer's website, along with the latest drivers, it is Dell IO Manager that lies, and HP IO Accelerator Management Server can be found on the HP website, respectively. It is worth considering that at the time of testing the driver on the HP website was version 3.1.1, which was listed on the official website as Private Version (I understand something like the Beta version), and Current Stable was 2.3.10
Well, as a starting point for testing, I will give the performance of our RAID10 on 6xSAS 146GB 15k:
Well, now the performance of two sections of our experimental ioDrive Duo
We were extremely surprised at such a low disk performance. Rolling back and reinstalling the drivers yielded no results! What does the disk showed the same performance in all modes of Maximum Write Performance and Maximum Capacity. We already had time to think that there might be a problem with the PCI-E port on our Dell server, so we decided to replace the test platform. Important factors for replacing steel: HP sticker on the disk itself, as well as more recent drivers on the HP website, which are not available to the mere mortals on the developer's website. Our new test platform has become familiar with comparative tests - HP DL380 G7 , in the following configuration:
- Processor - Xeon E5606
- Memory - 4 GB
- Hard drives - 2xSAS 146GB 10k
To begin with, the same HD Tune test was launched to check if there is any difference ...
And we saw her!
The difference is almost 2 times! That's what drivers can do. This is how it looks in the diagram:
Please note that the total disk capacity is 512GB, that is, in the Maximum Write Performance configuration, it “eats away” not only 50%, but only 20%, which is also very good. Of course, the failures on one of the sections do not look very good, but there's nothing to be done about this - the lack of support for TRIM makes itself felt. Now we are convinced that everything is in order with the disk, and we can proceed directly to testing, dropping unnecessary conversations.
Synthetic tests
AIDA64 Disk Benchmark
Crystaldiskmark
AS SSD
Of course, up to the speeds declared by the manufacturer - practically, as before the moon, but we have long been accustomed to the fact that the marketing figures still quite strongly disagree with the realities.
In the near future, we plan to conduct comparative testing of several SSD drives, including both conventional SATA and RAID, and PCI-E from other market players, and we will compare all of this with arrays from conventional SAS 10 / 15k disks. Moreover, many horror stories about the fact that SSD drives are unreliable, have long been transferred to the myths section, and they work no less than ordinary HDDs. Given the warranty period, you can safely rely on these products, but you should not forget about backups anyway!
By KorP
Source: https://habr.com/ru/post/160413/
All Articles