Also, what do u mean when u said that you find you can usually 'sneak by' with encryption?
Most of the time I'm 'hogging' bandwidth, it's to torrent files back & forth - usually on the order of a gig or three; if roadrunner 'sees' the multi-connect for simultaneous fast links, they throttle... If I encrypt the data going out, they never 'see' the data coming in - just looks like someone with three dozen web pages loading...
As I said, don't know much about trading data streams - but need to learn! Have in mind to do some 'trading stations'; think I can do real-time Black–Scholes machines for between 12 & 15 grand - depending upon options, and without the monitor array - "semi-supercomputer": GB UD9 w/6-core i7, 24G RAM, nVidia NF200 to run OpenCL on three 2G 5870's, one of them will handle six monitors, the rest just 'crunch', Areca 1680iX RAID card w/four SSDs in RAID0, and six 2G RE4's in RAID6 for 8G backed-up storage; "pretty much a supercomputer": same physical setup with Tyan 7025 MOBO & dual-6-core Xeons, dual IOH, for
full four slot PCIeX16 access to the 'crunchers'. Was thinking about nVidia Fermis, but you'd need a small nuclear power station to run the things, and the vidcard versions only do two monitor each - too limiting.
Short stroking:
Here's a picture of some 'drive guts' - with a ruler layed on top
...the little 'arm' carries the actual record/read heads, and it is 'driven in and out', from track to track, by a geared stepper motor. There are two physical aspects to accessing data: 1, the time it takes for the little arm to 'stroke' from where it's at, to where it needs to be, and 2, the speed of the magnetic material on the platter it's reading. In the picture, for guesstimates, you can see that the data tracks start at about three-quarters of an inch out on the platter, and end at about an inch and three quarters; the thing spins at a constant speed (10,000RPM in the case of a VR), so you can 'intuit' that, if the heads are at the outer edge of the platter, a lot more 'stuff' is flying under them, than when they are in toward the center - 1.75/.75, or about 230% more! If you simply don't partition the inner portion of the drive, a - the heads never have to make a full stroke, which cuts access time; and b, the data is restricted to the faster, outer area... Here's a graph of my data partition (on a pair of RAID1 WD RE3's, using the whole drives):
The left side of the graph is the 'outer edge' of the platters; because of the way the data is 'staggered' onto the drive, the effect is nowhere near the 230%, but you can see there is a two to one ratio, roughly, of difference between the two extremes... So, in its simplest form, short-stroking is simply partitioning only as much of the drive as you really need, and leaving the rest empty, to cut down on access time, and speed up transfer. I actually use a bit more sophisticated system, as I multi-boot a lot of OS's; my boot manager writes a master boot record to the drive every time I boot, which 'rearranges and makes visible/invisible' different partions each time - but the effect is the same - for my system and swap partitions, only a fraction of my drives are partitioned...
Your RAM is most likely running at 1333. RAM has a sort of little 'spreadsheet' chip on it, called an SPD, that contains the timing parameters to allow the board to automatically set it up - but the three standard 'columns' in the DDR3 'spreadsheet' are for 800, 1066, and 1333 speeds. Most faster RAM, yours included, contains another 'extended' data set, called either an XMP (Intel's standard - yours is XMP) or EPP (AMD/nVidia/Corsair standard) which allows a function in the BIOS to set up that faster speed... Yours will run considerably faster - the standard 1333 timings are 9-9-9-24, the XMP is 8-8-8-24 - roughly 20% faster. Wouldn't worry too much about it though, iCores are not very responsive to raw memory speed - never hurts, but not really noticeable in average use - YMMV!
Overclocking is easy, just don't want to go nuts. Most every Intel chip will do 20-25% over rated speed with very small increments of voltage increase, which, in the long run, is what kills chips. I do parameter sets for folks here, and never do anything that I think will have any even slight chance of reducing either the reliability, or long-term durability. The only caveat, is that I always recommend some (any) aftermarket CPU cooling. People go nuts with the cooling thing, buy stuff that's
way too expensive, when the fact is - even the cheapest aftermarket will have at the very least, two to three times the heat moving ability if the stock Intel 'rotary postage stamps'! I often recommend these, as they're reasonable, solid, and will fit nearly anything:
http://www.newegg.com/Product/Product.aspx?Item=N82E16835186134
but am doing a build next week for someone using one of these:
http://www.newegg.com/Product/Product.aspx?Item=N82E16835150113
to take a 3.2 i5-650 Clarkdale to at least 3.8, and likely 4.0GHz, assuming the thermals and memory interface look good (the Clarkdales are a little slow in the memory controller dept., as the controller itself is on a seperate 45nm graphics die from the main 32nm processor die...)
If you'd like a bit more info on the SPD thing, I'm working on a new sticky here:
http://www.tomshardware.com/forum/274400-30-memory-part-tweaking-tuning