by Andrey Kuznetcov
11/29/2004 | 11:32 PM
Contemporary sets of core logic differ from their predecessors by high level of integration, featuring all sorts of controllers onboard, which very often saves the system owners a lot of time and trouble acquiring and installing additional devices and expansion cards.
In our today’s article we are going to discuss a few technical solutions, which allow creating SATA RAID arrays on your home system without any specific efforts. The new HDD interface specification developed a while ago stimulated the appearance of the corresponding new hard disk drives in the storage market and pushed the chipset developers to implement their support in the new mainboard core logic sets.
The reason for this “adequate” response of the industry is certainly natural: the upcoming hard disk drives designed according to the Serial ATA standard will occupy more and more of the storage market, thus ousting the morally outdated PATA solutions. All large chipset developers have solutions among their products, which not just support Serial ATA hard disk drives, but also allow building RAID arrays from them. Here I am talking about chipset South Bridges developed by Intel, VIA and SiS.
And today we would like to test the four solutions from them in great detail.
We will evaluate the efficiency of the considered built-in SATA RAID controllers according to the performance of the HDD RAID array built with them. For this purpose we created RAID 0 and RAID 1 arrays of 2 Maxtor DiamondMax Plus 9 6Y120M0 hard disk drives with 120GB storage capacity each.
Here is the list of software applications we used for our test session:
Our test system was configured as follows:
With the help of a DataBase pattern we measured the HDD performance during processing of requests with different queue depth and different share of writes among them. During the test session we used 8KB data blocks with random address.
To make it easier to analyze and discuss the obtained results we built diagrams for queue equal to 1, 16 and 256 requests. First of all we will consider the controllers performance in case of RAID 0 array:
As always let’s start with the queue depth=1. The graph above indicates that VIA VT8237 is the most efficient of all the testing participants. It outperforms the opponents under all types of workload. It only yielded to SiS964 in RandomWrite mode. Both Intel controllers get kind of defeated by their Taiwanese competitors. However, the newer ICH6R controller is faster than its ICH5R predecessor, especially when it comes to processing write requests.
The increase in the requests queue depth up to 16 appears more efficient for Intel’s controllers. ICH6R retains leadership until the 80% of writes. The second Intel controller appears better than SiS and VIA solutions until we reach 50% writes share. The performance of VIA VT8237 and SiS964 grows up tangibly as the share of writes increases, but the first controller of the two still looks somewhat better.
In case we have a requests queue of maximum depth, Intel controllers strengthen their positions. The indisputable winner here is ICH6R, which yielded only to SiS964 in case of all writes. ICH5R runs somewhat slower than its newer fellow and starts losing to it more noticeably when the writes share grows as big as 70%. The above mentioned SiS964 gets ever more efficient as the share of writes grows up, so that in the end it appears the winner. The VIA VT8237 solution turns o9ut kind of weak here, although its lag behind SiS964 is noticeable only when the share of write requests hits 70%.
Now let’s take a look at the performance of our testing participants in the RAID 1 configurations.
The first graph shows the results obtained under linear workload. All integral parameters indicate Intel ICH6R as a winner here. Until the share of write requests hits 80% the Intel solution manages to outpace all other controllers. The second Intel controller is faster than SiS964 and VIA VT8237 in case of smaller share of writes among the processed requests, but as this number increases, its performance drops down. As a result, the second prize is won by SiS964, which manages to reach its maximum performance with 90% writes. VIA solution is running right behind it all the time.
With 16 requests queue depth both Intel controllers show smaller performance results as the share of writes increases. VIA and SiS solutions perform moderately but more stably proving efficient independent of the share of writes. The fastest solution here is Intel ICH6R. While SiS964 looks overall more attractive than the VIA VT8237.
The last graph displays the results for 256 requests queue. We see that both Intel fellows run almost equally fast here, ICH6R being slightly ahead. VIA and SiS products also proved very close to one another, SiS964 being maybe a little bit better. However, they are still quite far behind Intel’s solutions, managing to outpace them only in case of big writes share.
In order to provide you with extensive testing results, we also checked how fast they can handle sequential reading and writing when the array processes ordered requests with linearly growing address. The results are provided in the tables below and on 4 graphs built for both types of RAID arrays considered.
Let’s start with reading:
In RAID 0 array we see that VIA VT8237 controller appears better than the rivals when we process small data blocks. This advantage is achieved while the CPU bears really intensive workload, but this makes VIA solution the winner anyway. Both Intel controllers perform almost equally fast, while SiS964 appeared the slowest of all.
The linear reading graph for RAID 1 is very similar to the previous one. Again VIA VT8237 dashes forward during smaller data blocks processing. Intel’s solutions are again on the second and third place being farther ahead SiS964 due to their better performance with smaller data blocks.
Now let’s take a look at writing:
In case of RAID 0 the laurels are won by Intel controllers: this victory is achieved due to fast linear write speed during 2KB-32KB data blocks processing. In all other situations the results of the controllers are almost identical. The remaining two solutions are also almost equally efficient here, with VIA VT8237 being just a tiny bit ahead.
Another graph shows the linear write speed for RAID 1 array. I would say that here Intel controllers can hardly boast an advantage, because they are just a little bit faster when working with small data blocks than VIA VT8237 and SiS964. Among the other two controllers SiS964 appears somewhat better.
With the help of this pattern we tested our controllers under workload typical of a workstation with the maximum requests queue of 32 requests.
Let’s see how the controllers affect the RAID 0 array performance in this test:
On the first diagram you see how fast the array is when the entire storage capacity is used. The performance difference depending on the controller used is very evident. The best choice for a system like that would be Intel ICH6R. The second fastest solution is Intel ICH5R, you can see that the lines on the graph denoting their performance are almost parallel. We can see the advantage of the Intel products over the VIA and SiS competitors starting with the queue=2. VIA VT8237 is more efficient than SiS964 here, because the latter manages to show any performance growth only when we reach queue=16.
In order to evaluate the overall efficiency of the tested controllers we calculated the overall performance ratings for them according to the following formula:
Performance = Total I/O (queue=1)/1 + Total I/O (queue=2)/2 + Total I/O (queue=4)/4 + Total I/O (queue=8)/8 + Total I/O (queue=16)/16 + Total I/O (queue=32)/32
The performance ratings given on the diagram allow us to award the second prize to VIA VT8237 controller. It outpaces Intel ICH5R due to high relative role of its performance with queue=1.
Now let’s check how the controllers affect the results when only 32GB of the disk storage space are involved.
The diagram above shows that Intel controllers got even farther ahead, while the difference within the “pairs” (ICH6R-ICH5R and VIA VT8237-SiS964) got smaller. Nevertheless, the winner here will be ICH6R, the second fastest is ICH5R and the third – VIA VT8237.
Now let’s check out the performance ratings:
Here the visual evaluation of the results shown on the previous graph is completely proven by the overall rating diagram.
Now we would like to take a closer look at the results obtained in RAID 1 array:
The tests involving the entire array make Intel controllers faster than their rivals from VIA and SiS. ICH6R in its turn outperforms ICH5R. SiS964 and VIA VT8237 run almost equally fast.
The rating diagram shows exactly the same results as we have just seen on the previous graph. ICH6R is the most efficient of all, then follows ICH5R, and in the end VIA VT8237 is slightly ahead of SiS964.
Let’s consider a specific situation when only 32GB of the array storage space are active.
In fact there is no big difference compared with the results we obtained for the entire array storage space. Both Intel solutions are notably faster than the competitors, which run neck and neck. ICH6R is the No.1 here.
The performance ratings we calculated for the testing participants once again prove our point. ICH6R is an indisputable leader here.
Here we will take a look at the performance or RAID 0 and 1 arrays built with four different controllers and loaded as if they were typical file and web-servers.
Let’s start with the RAID 0 array working in the File-Server pattern.
On the diagram below we see that starting from queue=4 Intel controllers manage to get significantly ahead of the competitors. The best here is Intel ICH6R. SiS964 and VIA VT8237 do not yield to their opponents under linear workload, but then they get significantly more efficient when the queue grows up to 16 requests. VIA VT8237 controller is a little bit faster than SiS964, but as the number of requests in the queue grows up the advantage diminishes.
On this diagram we see the performance ratings for the tested controllers calculated as an average requests processing speed for different types of workload. Here we can clearly see how far ahead Intel’s solutions actually are. As for the performance differences between SiS964 and VIA VT8237, they are running really close to one another, the latter being just a little bit faster all the time.
When we have a RAID 1 array working in File-Server pattern we once again see how greatly Intel controllers outperform those from SiS and VIA. This advantage becomes evident starting with queue=4. Again ICH6R is faster than its predecessor. And SiS964 goes evenly with VIA VT8237 all the time.
The performance rating diagram shows how far ahead Intel controllers are and indicates equal efficiency of SiS and VIA products under this type of workload.
Now let’s take a look at the performance of these four chipset controller in Web-Server pattern.
On the RIAD 0 diagram we see a traditional situation with Intel controllers being ahead of their competitors and demonstrating a significant performance breakthrough starting from the requests queue=4. VIA and SiS solutions are almost equally fast here and they start picking up in speed only from 16 requests queue. The winner here is Intel ICH6R, which turned out just slightly ahead of ICH5R.
When we calculated the performance ratings for our testing participants, we again revealed an indisputable advantage of the Intel solutions. The performance difference between them and the products from VIA and SiS is almost 1.5 times. As for the VIA and SiS solutions, they again proved to be almost equally efficient.
The results for RAID 1 array show that Intel controllers retain their convincing leadership. The most efficient one of them is Intel ICH6R, which turns out just a little faster than ICH5R. The solutions from SiS and VIA demonstrate about the same level of performance under all types of workload.
The performance ratings for RAID 1 array indicate that Intel controllers are over 1.5 times faster than their rivals.
Having tested our controllers in a number of illustrative synthetic benchmarks we would like to spend some time discussing the results all four testing participants showed in the classical Winbench99 test. Disk WinMark numbers in the final tables are highlighted with different colors for your convenience: the highest scores are highlighted blue, and the lowest – red.
Let’s find out how the integrated chipset controllers affect the performance of a RAID 0 array first. We will start with Fat32 file system:
We can see on the diagram that the most efficient here is SiS964 controller. Hereinafter we will estimate the efficiency of our testing participants according to High-End Disk WinMark value. Moreover, the advantage shows by SiS964 solution over the competitors is quite significant, which is also true for the Business Disk WinMark result. The second fastest according to this test is the Intel ICH6R, which looks more attractive than its ICH5R fellow. VIA VT8237 controller appeared the slowest of all.
When we limit the storage space to 32GB, we do not see any great difference in the overall picture. Again SiS964 is far ahead of all, followed by ICh6R, ICH5R and VIA VT8237.
Now let’s take a look at the RAID 1 array results.
Again we see that SiS964 doesn’t give up its leading position. The rest of the controllers also retain positions in this race.
When we reduce the tested storage space to first 32GB of the array capacity, nothing actually changes. SiS964 remains the leader, although this time its High-End Disk WinMark scores got lower, unlike other controllers. The second fastest in this test is Intel ICH6R, and the third place is now successfully won by VIA VT8237, which turned out faster than ICH5R solution.
Having discussed the results obtained in FAT32 in great detail, we suggest paying due attention to another file system: NTFS, which gets more and more popular among the users today..
We will start with the full size RAID 0 array:
The diagram shows a totally different picture compared with what we have just seen in FAT32. The indisputable leadership stays with the Intel ICH6R controller, which appears considerably faster than any of the remaining three products. The former leader, SiS964, is now the slowest, while VIA VT8237 won the second and Intel ICH5R the third prizes.
When we reduce the testing area to 32GB we see that Intel ICH6R controller doesn’t want to give up the first position. It remains the leader having got considerably far ahead of the competitors. The second fastest is SiS964, followed closely by VIA VT8237. The last one here is Intel ICH5R.
How do our controllers affect the performance of a RAID 1 array? Let’s find it out now:
When we test the complete array, the victory is won by Intel ICH6R controller. Then comes VIA VT8237. the third prize is taken by the solution from SiS, while the very last one is Intel ICH5R.
In a 32GB storage space Intel ICH6R still feels quite at home remaining the leader. The second, third and fourth prizes are taken by ICh5R, SiS964 and VIA VT8237 respectively.
In conclusion we would like to present the results obtained in our FC-Test, which examines the arrays performance in real-time tasks and allows making some conclusions about the actual controllers efficiency. We used our traditional approach and tested the controller in five patterns, differing from one another by the average size and number of the files they consist of. All in all we performed four major operations. First we created files on the tested array in the first 32GB of the array storage space (Create). Then the set of files was read from the drive (Read). After that the set of files was copied within a single partition (Copy near) and in the end the set of files was copied from the first 32GB partition to the second one of the same size (Copy far).
Let’s first look at the results obtained for RAID 0 array type:
First of all let’s turn to FAT32 file system. The diagram made for Create pattern shows that at first VIA VT8237 controller appears the most efficient for work with a limited number of large files, however, in two other cases when we have a lot of smaller files involved (Windows and programs patterns) it turns out slower than we expected. In this case Intel ICH6R appears better performing. Also I would like to point out that SiS964 controller proved extremely stable: in four patterns out of five it remained the second fastest solution.
Read pattern doesn’t let any of the testing participants win. In Windows and Programs Intel ICH6R showed the best results, while in all other cases it was defeated by VIA VT8237 and SiS964.
When we copy files within the same partition Intel ICH6R and ICH5R show the best result. The slowest one in this case appears SiS964 controller, although I assume that the extremely low result in MP3 pattern can be explained by some driver issue.
When we investigate the copying of files from one partition to another the laurels get sent to Intel ICH6R controller, as it turns out the most efficient solution in four patterns out of five. VIA VT8237 and SiS964 move more or less equally, while Intel ICH5R loses to everybody.
The use of NTFS files system is more favorable for the Intel ICH6R controller, which wins in four patterns. The second position is fairly occupied by VIA VT8237. Then comes SiS964. ICH5R looks the weakest of all, as it was completely ruined by its opponents.
The reading results move VIA VT8237 into the leader’s position, as it managed to win in all five patterns. The second place is surprisingly won by the ICH5R controller. The third and the fourth positions were taken by Intel ICH6R and SiS964 controllers respectively.
When the files are copied within one partition, we see a very significant advantage of the Intel ICH6R controller. The gap is especially huge in the patterns consisting of many smaller files (Windows and Programs). Other controllers appear almost equally efficient in this test.
The file copying from one partition to another makes Intel ICH6R the winner. Although unlike the previous case all other solutions can also be “diversified”. The second fastest, as you may see, is VIA VT8237. The third position can be assigned to SiS964, which outperformed Intel ICH5R in four patterns out of five.
Now let’s check out the controllers performance in case we have a RAID 1 array.
In FAT32 file system we cannot actually name the leader. In Windows and Programs patterns Intel ICh6R is the best, while in all other patterns VIA VT8237 definitely looks better. SiS964 also demonstrated pretty good results, as it outperformed ICH5R and fell just a tiny bit behind the leaders.
File reading also doesn’t allow us to say who the winner is. The only thing we can state with all certainty is the fact that Intel’s duet is definitely the most attractive here.
During file copy within the same partition VIA VT8237 is the winner in ISO, MP3 and Install patterns. Also SiS964 looks pretty convincing. Intel controllers manage to keep pace with their competitors in all patterns except Windows.
The diagram built for file copying between the two different partitions allows stating evident victory VIA and SiS controllers over Intel’s products, which fall quite noticeably behind the leaders in most patterns. The performance difference becomes especially dramatic when we work with a limited number of large files.
The next diagram shows how efficient the tested controllers are when we perform Create operations in an NTFS formatted disk space. Here Intel ICH6R is the winner. It managed to cope best of all with a big amount of smaller files. Three other controllers are almost equally efficient, so that it is really hard for us to single out a winner among them.
File reading is best of all performed by the VIA VT8237: it appeared to be the fastest in all patterns. The second efficient is Intel ICH5R closely followed by its ICH6R relative. The last one is SiS964, though it is not falling too far behind the three leaders.
When we copy large files within the same partition, VIA VT8237 and SiS964 controllers show their best, however in Windows and programs patterns the situation turns to just the opposite. When we have to work with a lot of smaller files, the best solutions appears Intel ICH6R.
Copying files from one section to another names VIA VT8237 the winner in ISO, MP3 and Install patterns. In this case the VIA controller proves even faster than SiS964. In the remaining two patterns Intel ICH6R is better, as it manages to copy many smaller files much faster.
So we carried out detailed testing of the SATA RAID controllers integrated into the chipset South Bridges. The obtained results allow drawing certain conclusions about the strengths and bottlenecks of each testing participant.
The controller integrated into Intel ICH6R South Bridge turned out an indisputable leader in Intel IOMeter server tests. Its advantage over the rivals grows bigger as the number of requests in the queue increases. Even despite a more powerful test platform, which still has some influence on the final benchmark results, we would still claim that using Intel ICH6R controller within a SATA RAID array of a server disk subsystem makes perfect sense. It is true for both: RAID 0 and RAID 1. As the share of writes among the processed requests grows bigger, the controller loses its efficiency. We can also say that this controller is an indisputable leader in WinBench 99 tests, when the drives are formatted in NTFS file system. The performance of the RAID array is much higher than in case of FAT32. This may be the result of controller driver optimization for NTFS file system, as the latter gets more and more popular nowadays. The indirect proof to this point will be the results obtained in our home FC-Test. You can see that Intel ICH6R works very fast with patterns consisting of many smaller files in NTFS (unlike FAT32). The best performance in FC-Test can be achieved in case of RAID 0 array. In case of RAID 1 Intel ICH6R slows down as it comes to copy operations.
The controller integrated into an older Intel ICH5R South Bridge yields to its newer fellow. In server Intel IOMeter benchmarks the gap between them is not that big, which allows us to claim that Intel ICH5R outperforms successfully VIA VT8237 and SiS964. This controller looks less attractive when it comes to RAID 0. As the share of writes grows up, it slows down quite noticeably. In WinBench99 tests ICH5R looks less attractive against the background of its newer counterpart. Its efficiency doesn’t seem to depend on the file system used, as well as by other integrated controllers. In FC-Test ICH5R is defeated by the newer Intel controller in most cases when we work with NTFS file system, especially when we have to process a lot of small files. At the same time we would like to point out that it hardly yields to the remaining two testing participants in this test.
SATA-RAID controller integrated into the SiS964 South Bridge falls significantly behind Intel’s solutions in Intel IOMeter tests and performs almost as fast as VIA VT8237. As the share of writes increases, it remains pretty stable. In FAT32 in WinBench99 tests it turns out the winner, outpacing all the competitors in both: RAID 0 and RAID 1 arrays. In FC-Test SiS964 appears faster in case of RAID 1, especially during file copy operations.
Finally, a few words about the last of the four controllers: the solution integrated into VIA VT8237 South Bridge. In Intel IOMeter benchmarks it is almost as fast as SiS964, as we have just said, and yields to Intel’s products. The increase in the writes share doesn’t actually affect the performance of this solution. In WinBench99 tests VIA VT8237 controller is almost as fast as two other products following the leader. It is true for both file systems and both array types. FC-Test shows that VIA VT8237 controller is slightly faster than the solution from SiS in most cases when we have to work with large files.