WHEN I FIRST saw the product that Adaptec is calling MaxIQ at CeBIT in March, it looked interesting, but was overwhelmed by the 5Z ultracapacitor backed RAID card. Now that Adaptec has let out all the details, the MaxIQ SSD RAID accelerator is by far the more interesting product.
The idea itself is simple. You take a fast SSD, in this case an Intel X25 SLC based 32GB drive, and use it as an intelligent cache for your raid array. SSDs have many things that make a drive great – speed, no penalty for random reads/writes, and low power. They are however currently lacking some things many consider to be deal breakers, which are capacity, price and standardization.
MaxIQ in action, note the scores
With MaxIQ, Adaptec is looking to build a best of both worlds device, something they are calling a High Performance Hybrid Array. If you take a standard Adaptec RAID card – any 2-, 5-, or 5Z- series part will work – and simply update the firmware, you can just plug in one of the special Adaptec branded drives. The caching system works automagically from there.
If you think of MaxIQ as a big cache for the RAID array, you are mostly right, but instead of sitting between the CPU and drives, it sits beside them, working in parallel. The software and algorithms on the RAID card are proprietary, but Adaptec gave SemiAccurate a few hints about how it works.
How the data flows in MaxIQ
When any sector is written to an array, it is considered ‘hot’. While the data is being written to the magnetic drive arrays, some or all of it is also written to the SSD in parallel. Because writes are frequently re-read, the most recently written data is considered hot and cached.
On writes, it will usually cache the first sector to the SSD, allowing it to be read in orders of magnitude less time, starting the data flow to the controller while the magnetic HDs are still seeking. If the cache tuning elves at Adaptec did their job right, the data on the SSD should be enough to mask the magnetic HD’s latency.
MaxIQ can cache anything that is written to the disk. It doesn’t necessarily care what app, data, or format you use, just how frequently. That may be an oversimplification, since there are likely layers and layers of rules to determine what is hot and what isn’t, but the general idea is correct. If it is used a lot, it is likely cached.
Data is stored on the SSD until it is evicted by a proprietary set of rules. The MaxIQ software also replaces the garbage collection on the SSD to make it less noticeable, hopefully never slowing down the caching as it can with some consumer SSDs. Since writes to the cache are relatively infrequent, this should not be a hugely complex problem to solve.
Why bother with a cache at all? There are a few answers to that question, and the first and most obvious is raw performance. Using a single SSD as a cache, Adaptec can squeeze much more performance out of an existing array for a relatively low cost. That part is fairly obvious.
More important is the other reason, which is kind of a make or break problem for Adaptec. SCSI, SAS, and SATA adapters, even higher level RAID cards, are somewhat commodity now, so there is little magic left. Adaptec is in the enviable position of being in the middle of the data stream between storage and compute, since almost all data used by a server passes through its cards if it is coming from local storage.
How Adaptec sees the world
The problems of reliability and scalability, at least for anything that is attached to a single server, have long been conquered, so what do you do to remain relevant if you make RAID cards? Add smarts to that data path to add features and speed things up at a price/performance level that nothing else can touch. Since Adaptec is already in the data path, MaxIQ can bring a fair bit of SSD performance to magnetic storage arrays without the full SSD price tag.
Since the MaxIQ drive operates in parallel with the magnetic array, none of the existing data protection is modified or compromised, data I/O just goes faster. If the power goes out, the normal RAID card cache does what it always does, and the data on the SSD simply sits there. It is totally transparent to the user, so anything that works with a normal Adaptec card should work faster with MaxIQ.
There is no setup, no configuration, and no user tweaking needed, you just plug in one to four SSDs and they just start working. If the RAID card you are using has the correct BIOS, you can even hot-plug it, but read the manuals carefully before you take my word on this, because data recovery on a RAID array is a royal pain.
The smarts for MaxIQ are on the card. The drives themselves are modified Intel X25 SSDs that have a special signature that the Adaptec cards see. If you plug them in to a normal computer, they work like a standard X25, which is to say very well. A ‘normal’ X25 plugged in to a 2- or 5- series card will behave just like a plain X25; you need that signature before it becomes a cache.
Before you get ideas about dd’ing a drive with a signature, we are told that it resides in an area not normally accessible, so cloning the drive won’t buy you anything. If you want the caching, you need to buy the Adaptec branded cache units. That said, they are one of the easiest ways out there to speed up an array, so it is money well spent.
How fast is it? Adaptec is claiming up to 5x the IOPS of any similarly priced solution, and much lower power use on top of it. The specific numbers that they gave out were a 5x increase in web server loads, 4x in fileserver work and 4x in OLTP type scenarios. In raw random reads, Adaptec claims 11x more speed.
These numbers are from an array with 12 15K Seagate SAS drives backed by a single SSD cache. Adaptec says that it can get up to 20K IOPS and 1.2GBps throughput with the MaxIQ system. If you check its web site in the coming days, there will likely be some white papers and a lot more specific numbers, there always are for this kind of enterprise product.
In the end, if you have an Adaptec array, MaxIQ seems like a no-brainer upgrade, and it is worth a long look if you are still deciding what to buy. The price tag of $1295 per drive may seem a little steep by enthusiast PC standards, but on the enterprise side, it is not expensive at all.
When you consider the installation costs, almost zero if your BIOS is up to date, the configuration costs, zero, the maintenance costs, zero, and the added power costs, almost zero, the ROI seems quite compelling. Add in a potentially huge performance increase, coupled with removing a perennial headache in array latency, and there is almost no downside to MaxIQ.
Until SSDs drop in price per GB to a small multiple of magnetic storage, hybrid schemes like this seem to be the way forward. MaxIQ is the first one out of the gate, and it seems to be done right. You can hardly ask for more.S|A
Latest posts by Charlie Demerjian (see all)
- Thing go bump(gate) in the night for Nvidia’s GP100 Pascal GPU - May 3, 2016
- Cavium’s Octeon TX blends compute and packet moving - May 2, 2016
- Qualcomm releases Zeroth API to developers - May 2, 2016
- Another detail about Qualcomm server SoCs revealed - Apr 27, 2016
- AMD finally really honestly launches the dual Fury - Apr 26, 2016