Graph would look better ...
... if the Y axis wasn't "lower = better" . Wouldn't the ("bang per buck") reciprocal create less cognitive dissonance on a Monday morning?
An all-flash 3PAR array has stormed to the top of the SPC-2 benchmark charts. Nice one HPE, you've timed it just right for your independence from PCs and printers. The 3PAR 20850 scored 62,844 aggregate MB/sec, topping EMC’s VMAX 400K and its 55,643.78 MB/sec. This benchmark measure array performance in terms of throughput ( …
For those not willing or able to do the maths...62,844 *8 (to Mbps) /1024 (gbps) /16Gbps (per FC) = 30.69 ports. With overhead that's 32 front end ports fully saturated.
The article essentially says that the 3Par is able to fully saturate all of its FC ports, just like every other SAN on the market. This one has more usable 16Gb FC ports (32) than previous SANs tested, and costs less than previous SANs tested. Any future flash based SAN with more usable bandwidth will be faster than this once tested (because flash can saturate any throughput) and any future SAN tested which costs less will also beat this.
Interestingly, because the second socket is taken up with an ASIC rather than a programmable CPU (Xeon) the 3Par architecture is limited on PCIe bandwidth per controller and so is theoretically trivial to outperform controller for controller when compared with competition. I would expect a NetApp 8080 cluster to trounce this, for instance, if they were to do a test. Not that it matters for any normal consumer, I always struggle to work out who these tests are aimed at, especially now the limitation is essentially PCIe bus in the controller rather than disk or controller performance.
They used 96 ports and 8 switches(it's in the disclosure doc).
Scales to 160 16gig FC.
Rated for 75gig a sec (read). Some of the SPC2 tests showed higher throughput but it fell relatively short(er) on the large file test. So overall 63gig. Probably 3-5x what any earlier gen 3par could do.
First SPC2 ever for 3PAR i believe.
The system will go faster in the future as the software matures to the new hardware (talked about this during discover)
You're right, they actually had 96 FC ports in the controllers, how come they were unable to saturate these given the massive SSD and SAS capability behind them? Using flash I always manage to saturate the front end even on longer tests so in a test designed around this end I don't see how they got such an apparently poor result.
I'm also surprised that the primary use-case HP came up with is VOD. Seems a very expensive way to do video to me when NL-SAS has a very similar sequential read speed to SSD but with massively superior cost/TB
RE: "I would expect a NetApp 8080 cluster to trounce this...."
I wouldn't be so confident. NetApp have never submitted a FAS based SPC-2 result. Their first ever SPC-2 result was submitted in Sep/15 on the E-Series @ 8,236 MBPS. If they could submit a trouncing result .... my experience in this stuff suggest to me they would!
/Paul Haverfield [HPE Storage]
>The total list price was $3.2m, discounted to $1.25m
What the fuck is this bullshit. If even a single customer does not get that discount then surely its position on that graph is invalid.
So, is it a "lets use some really low figure that we give to that giant customer who uses nothing but HP kit" type number?
If everyone gets that price then why do they even mention $3.2m?
Standard storage pricing practice. If your supplier has not explained this to you, your supplier is ripping you off.
The "list" price is the headline figure used in order to be taken seriously against the IBMs and Hitachis of the world and there is never an intention of anyone paying this. If you don't get at least 40% off list you've done something odd. The difference between 40% off and a percentage I won't share with you but which I regularly achieve is how well you know the game. It's an easy game, but one with various rules that any good supplier will know. You, the customer, also need to be a part of this game. Again, if your supplier didn't tell you the rules and how to play the game before you ask for quotes you need to change supplier.
Now, go find a supplier with someone called Anonymous Coward on their books...
Hah. Like I get a say in purchasing.
But if you get a 40% discount on this product and some much much larger customer gets a 61% discount like they mention then you can see that you would be annoyed that the graph shows them down there in the bottom right while you are higher along the Y axis .... Your price performance is significantly lower than theirs for the same product.
Like I said, bullshit graph.
I have a friend who has worked in EMC for a long time. I remember him saying one time that the only customer ever charged full list price for one of their mainframes was EMC. I'd imagine if some EMC buyer rang up HPE asking for prices of their fancy new all flash box they would not get much of a discount, more likely the reverse.
AFAIR, according to SPC rules, discount provided by vendor in test must be granted to anyone who decided to buy the same config. BTW, EMC VMAX has the same discount.
Moreover, HPE has FC switches for $150K after 50% discount while VMAX connected directly.
I heard several anecdotal stories about buying on list price, for instance, when one vendor wanted to buy competitor's equipment for it's lab:)
"But if you get a 40% discount on this product and some much much larger customer gets a 61% discount"
40% is the absolute minimum you should expect. 61% is better. Nobody on these forums will tell you the maximum you can get, that's down to your negotiation skills. If you don't get 61% off list for your enterprise storage change your supplier. Bear in mind though that they might have just never showed you the list price :)
62% off list is not extreme by any stretch for HP 3par. After HP boufht 3par one of the first things they did was jack up pricing so they could show higher discounts. I heard part of it was agreements with aome distributors that required certain levwl of discounting. Prior to that 3par regularly walked away from deals that needed more than i think 47% discount.