Sign in with
Sign up | Sign in

Controllers And Setup: Adaptec RAID 5805

Breaking Records With SSDs: 16 Intel X25-Es Do 2.2 GB/s
By

Adaptec’s state-of-the-art Unified Serial host adapter product line is called the 5-series. There are many different models, each tailored to different internal/external storage requirements, such as low profile versus full-height cards as well as various port counts. Model 5805 is a low profile, eight-port internal SAS/SATA card.

Adaptec currently offers up to 28 ports on a single card. However, we deliberately went after two eight-port cards instead of one card with a massive number of ports so we could distribute the bandwidth across two PCI Express slots. If you look at the typical SAS/SATA HBA interface, you’ll find that it’s a first-gen x8 PCI Express connection, which reaches a maximum of 2 GB/s. Since we wanted to reach higher throughput, we had to go for two cards and create a software RAID array using the operating system.

We looked at the 5-series by Adaptec more than a year ago, but it’s still a top notch product line. The latest upgrade to the famnily was the 5Z-series, which introduces Adaptec’s Zero Maintenance Cache protection. Conventional RAID controllers come with a cache memory, typically ECC DRAM, and an optional battery backup unit, which maintains the cache content in the case of a power failure. Adaptec’s approach integrates flash memory with the controllers, saving the DRAM cache content into non-volatile flash memory.

System Details and Device Configuration

We decided not to assemble a purpose-built system for this project, since we wanted the 16-drive flash SSD array to be suitable to any upper-mainstream system, such as our Storage Reference Test System. Hence the test environment consisted of a Core i7-920 (2.66 GHz) on a Supermicro X8SAX motherboard, 3 GB of Corsair CM3X1024 DDR3 memory, and an OCZ EliteXstream 800W power supply.

The only change we had to make to the reference system configuration was the graphics card. The Radeon HD 3450 had to go away, as it utilized a PCI Express slot we needed for the RAID controller. The replacement was an old GeForce 4 MX 440 with 128 MB of memory and a PCI interface.

Display all 122 comments.
This thread is closed for comments
Top Comments
  • 13 Hide
    chise1 , July 30, 2009 6:25 AM
    can we have some benchmarks that aren't just I/O performance? How about boot times and/or program load times?
  • 12 Hide
    lutel , July 30, 2009 6:20 AM
    how fast does it open solitaire ?
Other Comments
  • 4 Hide
    xyz001 , July 30, 2009 6:11 AM
    how fast does it boot windows?
  • 6 Hide
    IronRyan21 , July 30, 2009 6:15 AM
    can toms give this away like the SBM! I have no idea why I would need this tho. :) 
  • 12 Hide
    lutel , July 30, 2009 6:20 AM
    how fast does it open solitaire ?
  • -8 Hide
    afrobacon , July 30, 2009 6:24 AM
    Porn delivered in .1 seconds or your (insert something witty) back...
  • 13 Hide
    chise1 , July 30, 2009 6:25 AM
    can we have some benchmarks that aren't just I/O performance? How about boot times and/or program load times?
  • 4 Hide
    dirtmountain , July 30, 2009 6:36 AM
    You should always include a retail price tag for these articles. If it's in there someplace i missed it.
  • 3 Hide
    apache_lives , July 30, 2009 6:58 AM
    Any non windows based benchmarks incase there is any sort of limit of throughput etc?

    Windows does some funky things to hdd transfers - buffering things through ram and all sorts to find extra performance - wouldnt supprise me if that 2gb/s limit had something to do with software accessing the ram through the layers and windows subsystem etc
  • -7 Hide
    falchard , July 30, 2009 6:59 AM
    I am pretty sure the new Intel SSDs still don't have a good write speed compared to the Indolex controlled SSDs.
  • 1 Hide
    apache_lives , July 30, 2009 6:59 AM
    xyz001how fast does it boot windows?


    half of the start up time on the windows side (aka not including bios time) is the PNP initialization and network loading/waiting etc - check the hdd read light on high end systems
  • 1 Hide
    apache_lives , July 30, 2009 7:00 AM
    falchardI am pretty sure the new Intel SSDs still don't have a good write speed compared to the Indolex controlled SSDs.


    Every other spec Intel owns hands down like random writes etc which makes them the far better drive
  • 5 Hide
    cangelini , July 30, 2009 7:25 AM
    dirtmountainYou should always include a retail price tag for these articles. If it's in there someplace i missed it.


    Dirt,
    You're looking at close to $14k worth of drives/controllers :) 
  • 3 Hide
    amnotanoobie , July 30, 2009 7:31 AM
    Too bad my money tree couldn't buy me even one X25-E.

    And yeah where are the application load times?
  • 0 Hide
    Ramar , July 30, 2009 7:57 AM
    When/if I ever have enough people paying me for space on my server, I know what to do.

    We've come a long way from "Loading..." screens in Half Life 2 every five minutes or less.
  • 5 Hide
    dean heart , July 30, 2009 8:01 AM
    Gonna say it as well: Please benchmark application loadtimes; photoshop with different filesizes and ofcourse level loadtimes in Crysis :) 
  • 4 Hide
    chyll2 , July 30, 2009 8:32 AM
    I wish they also have real-world results/benches. Im not that familiar with synthetic benchmarks.
  • 3 Hide
    al2950 , July 30, 2009 8:35 AM
    You will not be able to get faster speeds than that using 2 8x PCI-E. Even though the theoretical bandwidth is 2GB/s I have only even been able to get around 1.15GB/s, whwich is pretty close to what you are seeing. I would be interested to see what happens if you use 3 Raid controllers :) , although i cant remeber how many total physical lanes are available on the X58 chipset
  • -4 Hide
    profundido , July 30, 2009 8:36 AM
    Dear Tom,

    another great article! Logically the cpu power should be the bottleneck, therefore you should try loading up same the config on a dedicated dual or multi cpu servermotherboard with a windows 2008 Server R2 RC 64-bit as OS for more simultaneous cpu operations. That might bump up your figures beyond 2.3GB. And then finally, this is a bit "breaking the frontiers" but hey isn't this what you guys are known for by now...you should grab that new workstation board from Asus (forget the exact name) that's filled only with PCIE slots (about 5 or more I think) and try adding 2 more adaptec cards with each 4 SSD's. This would eliminate the possible bottleneck of limited cpu operations per raid controller.
  • -8 Hide
    ossie , July 30, 2009 8:38 AM
    "Bottlenecks can most likely be found in CPU performance as well as farther down the platform in the storage controllers."
    That's over-simplified, if not pure B$. Any modern CPU has more than enough BW. There are a lot of other limiting factors, as local buses, memory, and last, but not least, the OS (crappy vi$hta DRM-O$).
    As both arrays (the more heterogeneous one from Samsung and that one) are hitting a very similar peak transfer rate (5% doesn't really count), despite very different HW setups, the most probable explanation lies in the OS as limiting factor (the single common denominator).
    As for the retarded comments, inquiring windblow$ booting, or some application, or crappy game level loading times:
    A large RAID is nothing for desktops, with inherent weak task and IO parallelization, but for servers with high IOPs and a lot of clients.
  • -2 Hide
    mitch074 , July 30, 2009 8:53 AM
    I wonder what performance Linux's ext4 file system would get out of that array... Since, after all, Windows (any version) is sorely lagging behind *NIX systems on I/O throughput.
  • 3 Hide
    tacoslave , July 30, 2009 8:55 AM
    from what ive seen those are perfectly valid questions because we ARE reading because were curious. By the way most comments on toms arent retarded (flaming,fanboys = retard post.)Anyways I think most of us were thinking the same thing since most of us won't ever buy something like that. windows boot time = around 2 min for my pc
    ultimate array = ?
Display more comments