3Ware Escalade 7500 w/12-channel ATA RAID

thebs writes: “I just hit 3Ware [Homepage] looking for firmware/driver updates and they just announced a new ATA RAID product today. The new series is the 3Ware Escalade 7500 available in 4 (7500-4), 8 (7500-8) [Datasheet] and a new whopping 12 (7500-12) [DataSheet] channel ATA version! [ Small picture of the 12-channel version [Image] ]

It appears EETimes [Homepage] had already published info regarding the upcoming 3Ware Escalade 7500 series [Article] back on June 24th.

Quick Specs:

• 64-bit/33MHz PCI (266MBps)
• 64-bit ASIC (application specific integrated circuit)
• 0 Latency SRAM (static RAM, not synchronous dynamic RAM)
• 4, 8 and 12 channel ATA options (7500-4, 8, 12, respectively)
• Only one drive used per channel (no master/slave issues)
• ATA UltraDMA/133 Speed
• Full Linux support (3Ware-written GPL driver, in stock kernel since 2.2.15)
• Multiple RAID-5 volumes on 7500-12 product
• Also designed for upto 16-port SerialATA (with external PHY chips)

Further Discussion:

• How much memory is on-board?

  I cannot tell the amount of SRAM memory on these cards and 3Ware doesn’t
  market the size each product sports (for reasons that will become obvious in a bit
  to those new to 3Ware’s designs). The existing 6×00/6×10 and 7×00/7×10
  series have 1MB SRAM and the 7×50 series have 2MB SRAM. From the looks
  of it, there are two (2) SRAM ICs on the 7500-4/8, just like the 7×50,
  but four (4) SRAM ICs on the 7500-12. In fact, the 7500-4/8 looks like
  the exact same PCB (printed circuit boards) of the 7450/7850. So I’ll
  assume those are each 1MB, 32-bit SRAM chips on-board, which would equal
  2MB for the 7500-4/8 (just like the 7×50) and 4MB for the new 7500-12.

  Only 2-4MB? You thought most [true hardware] RAID cards had 16MB+,
  right? I mean, the Adaptec 2400A and Promise SuperTrak series sports
  upto 64MB, eh?

• What is SRAM? How does it differ from SDRAM? Advantages/Trade-offs?

  Understand that SRAM = static RAM, not Synchronous Dynamic RAM —
  big difference. SRAM is the logic used in cache memory, not main
  memory like SDRAM/RDRAM. While the burst write performance of SDRAM is
  similar in performance to SRAM, the random access, let alone any read
  operation, performance is almost an order of magnitude faster with
  SRAM. The key is latency. SDRAM is still DRAM. While it has a
  sub-10ns, synchronous burst write timing, it is still a 50ns+ (typically
  70ns+) memory technology for initial access, such as when reading.
  That’s why SRAM is used in cache logic, to overcome the huge latency
  hits when the CPU has to read from DRAM main memory. I won’t go into
  the electrical and design differences between SRAM and DRAM, just know a
  SRAM cell is much bigger than a DRAM cell (hence why you get a lot less
  for even more money).

  This is also why the 3Ware cards use an ASIC (application specific
  integrated circuit) instead of a microcontroller (i960, StrongARM,
  etc…) like the Adaptec 2400A, Promise SuperTrak series and most SCSI
  RAID cards. We’re not using buffered DRAM, but 0 latency SRAM that
  cannot be impeded by overhead and buffering. So it is often directly
  I/O mapped memory for transfers, which the 64-bit ASIC provides without
  delay. Using a microcontroller would negate the performance benefits of
  SRAM, since it cannot act like a simple bus arbitrator like an ASIC
  can. And to make matters worse for competing ATA solutions, the Adaptec
  2400A and Promise SuperTrak use very slow i960 microcontrollers (at
  least versus faster i960 and, even more so, StrongARM chips on typical SCSI RAID
  cards) — especially the SuperTrak. So their microcontrollers are the
  bottlenecks compared to even the 133MBps 32-bit/33MHz PCI bus
  (SuperTraks seem to be “stuck” at ~40MBps).

  The trade-off of the design is, of course, that the 3Ware cannot buffer
  as many RAID-5 XOR operations because of the vastly smaller amount of
  total RAM. So it’s quite likely to stall with a significant number of
  random RAID writes. Many argue, including myself, that most OSes flush
  their disk buffers/cache so writes are as linear and contiguous as they
  can be anyway. So it is debatable how much “worse” 3Ware cards are for
  RAID-5 volumes. It might have been an issue with older models that only
  had 1MB cache, but 2MB in the 7×50 and 7500-4/8 and, assuming I’m
  correct, 4MB in the 7500-12, should be enough to cover the majority of
  applications that incur a massive number of RAID-5 writes.

• More on “True hardware” ATA RAID v. “BIOS-only” ATA RAID

  For more on the differences between ATA RAID options, please see my
  “draft article” (I never finished finalizing it, so ignore the grammar issues) here entited “Dissecting ATA RAID Options” [Article]
  It was originally written for, and but picked up by, CMP’s “Sys Admin” magazine for their July 2002 issue on Storage.

• The future: 3Ware Escalade 7500 w/16-port SerialATA

  The next step in the 7500 series will be SerialATA. According to the
  EETimes article, upto 16 ports will be offered. I can only assume the
  new 3Ware Escalade 7500 has been designed with both legacy, parallel ATA
  and SerialATA in mind. SerialATA was designed so controller logic compatible with legacy, parallel ATA could be used designed, only requiring the addition of an external physical interface chip (PHY) on the board to support SerialATA. The EETimes article confirms this is the plan 3Ware has for the series, with a $100 premium over the legacy ATA version covering the cost of the addition of the PHY chips.

  For more on SerialATA, including design and engineering issues, please
  see my previous PC_Support post entitled “An introduction to SerialATA, the future of commodity storage” [List Post].

“