Are the wheels coming off?

From Term Sheet (required reading BTW)

Since landing in San Francisco on Wednesday afternoon, I’ve met with an assortment of senior venture capitalists, bankers, entrepreneurs and crossover investors. All of them have, in one way or another, been involved with so-called ‘unicorn’ companies. As in the past, they are nearly unanimous in sentiment. The difference now is that their sentiment is fear.

The past several years of raising too much, too high, too soon has run smack into a much more conservative investor ethos. Later-stage tech startups can still raise growth equity — and still lots of it — but not necessarily at the terms they were receiving just two months ago.

“This shift is only five or six weeks old, so most companies haven’t felt it yet,” a senior tech banker explains. “But I know of many companies who raised money at $1 billion valuations last year that are now being told that, to raise money now, they need to take around $700 million or $800 million. Probably with some serious structure that protects investors, like ratchets, on top of it.”

The reality is that record-high private market valuations have been driven by two things: Wall Street’s lust for growth at all costs, and relatively high tech multiples in the public markets (and, more specifically, applying the former to the latter). But a variety of macro economics factors (China, the inscrutable Fed, etc.) have cut public equity prices and moved the spotlight to unit economics, which means some pretty large biz model disruption for the disruptors.

Read it all.

The thing about bubble valuations and unicorns … neither one will last very long. Pure Storage IPOed this week and they aren’t doing as well in the public markets as their private market valuations might suggest. This is not to say they aren’t a good company, or don’t have a good product. This is saying that the demand for “unicorn” valuations from the buy side is … well … weak. The market values real businesses able to generate operating capital from revenue.

Is this the beginning of a (not so great) unwinding? I dunno.

I do think that this will encourage a) companies to sell, b) others to acquire, and have them all do it at more reasonable valuations.


Viewed 37 times by 35 viewers

possible M&A: Dell and EMC?

Story is here. Not sure this is a great tie up … EMC has lots of things Dell doesn’t need (and vice versa).

Possibly parts of EMC (secession from the federation?) with Dell. I can’t imagine VMware wanting to tie up with one vendor. Nor Pivotal, etc.

This said, Cisco pulled out of the venture with EMC to pursue its own directions, competitive with elements. But then they bought and subsequently closed Whiptail.

I am not really sure I know what to make of this. HP Enterprise with EMC storage would be leaner/meaner I think, though HP acquisitions haven’t been all that successful.

I dunno … maybe de-federating, and splitting the company up to pursue markets would be better, but I can’t see how it would impact much (other than to return capital to shareholders … which isn’t a terrible idea).

If they did split up, I could see the storage piece looking for a happy home for vertical integration. I could see DSSD, Isilon, etc. under one roof, though you’d have to be careful to prevent making regulators uncomfortable … so there would likely be more divestiture.

Viewed 1906 times by 307 viewers

End days must be on hand … Perl 6 is out

see for more details. I’d love to find a valid reason to play with it, but my near term foci are going to remain our current code base in Perl/C, nodejs for a few things, Julia/R for analysis.

The joke about Perl 6 shipping by Christmas is now over … as the correct response has been “what year”. Until this year it seems.

Viewed 3099 times by 384 viewers

As the benchmark cooks

We are involved in a fairly large benchmark for a potential customer. I won’t go into many specifics, though I should note that lots of our Unison units are involved.

Current architecture has 5 storage nodes (6th was temporarily removed to handle a customer issue). Each Unison node has a pair of 56GbE NICs, as well as our appliance OS, and bunches of other goodness (quite a bit of flash). Total capacity for test is of order 200TB of flash.

Looking at one of the Unison units in detail as it is cranking away:

I am really enjoying that with 6GB/s into the unit for storage purposes, our user load is hovering around 4.8. We are operating less than 60% of this units maximum practical storage speed here … the network is actually the bottleneck in this case.

Even more to this, I am enjoying watching the 8 hosts drive the system at 24+GB/s for something that is not quite a streaming load (think many small files).

These units (heck, this unit!) are available for purchase, and we are working on convincing this user that they really need to deploy this. Performance, real, actual, tangible, measurable performance … matters … and it makes a huge difference.

Lots of people quote marketing numbers for performance. We deliver the massive firepower required to provide the performance. There is a significant gap between promises and measurement. Worth noting when you start comparing “equivalent” systems.

Viewed 22581 times by 1018 viewers

Inventory to sell to make room: Cadence and several Unison/JackRabbits

Very fast units, very reasonable prices. We are (again) running out of space in our lab, and really need to move this stuff out. Many of these have been demo/engineering machines for us, including the portable petabyte unit.

We’ve got a Cadence box with 16TB of storage, which puts up performance numbers that other vendors would kill for …


We’ve got the portable petabyte unit available (albeit with less than 1 PB).

Let us know if you’d like one or more (or all!).

Viewed 25448 times by 1060 viewers

Updated net-tools bits

So far, 3 components, and working to fix a few things in formatting. On github, grab it here.

First, to report about bond details

root@unison-mgr-1:~/net-tools# ./ 
bond0:	mac 0c:c4:7a:48:69:cb
	state   up
	mode fault-tolerance (active-backup)
	xmit_hash layer2 0
	active slave eth1
	polling 100 ms
	up_delay 200 ms
	down_delay 200 ms

	slave nics:
	eth1: mac 0c:c4:7a:48:69:cb, link 1, state   up, speed    1000,
		driver igb, version
		firmware version 1.61,0x8000090e

bond1:	mac 00:12:c0:80:26:76
	state   up
	mode fault-tolerance (active-backup)
	xmit_hash layer2 0
	active slave eth3
	polling 100 ms
	up_delay 200 ms
	down_delay 200 ms

	slave nics:
	eth2: mac 00:12:c0:80:26:76, link 1, state   up, speed   10000,
		driver ixgbe, version 4.1.2
		firmware version 0x2b2c0001
	eth3: mac 00:12:c0:80:26:77, link 1, state   up, speed   10000,
		driver ixgbe, version 4.1.2
		firmware version 0x2b2c0001
	eth4: mac 00:12:c0:80:26:72, link 1, state   up, speed   10000,
		driver ixgbe, version 4.1.2
		firmware version 0x2b2c0001
	eth5: mac 00:12:c0:80:26:73, link 1, state   up, speed   10000,
		driver ixgbe, version 4.1.2
		firmware version 0x2b2c0001

Second, to show bridges components

root@unison-mgr-1:~/net-tools# ./ 
br10:	interfaces eth0
br20:	interfaces bond0,vnet0
br30:	interfaces bond1

and finally,, to (shortly) act as a {bridge,bond,ipv6} aware version of ifinfo (which it should replace).

root@unison-mgr-1:~/net-tools# ./ 
  DEVICE [  MASTER] LINK: IP Address                     MTU RX (bytes) TX (bytes)
   bond0 [    br20]   up:                               1500 71.769 MB 90.609 MB
   bond1 [    br30]   up:                               1500  0.001 MB 20.718 MB
    br10 [        ]   up:      1500  0.084 MB  0.217 MB
    br10 [        ]   up:  fe80::ec4:7aff:fe48:69ca/64  1500  0.084 MB  0.217 MB
    br20 [        ]   up:     1500 70.034 MB 65.839 MB
    br20 [        ]   up:     1500 70.034 MB 65.839 MB
    br20 [        ]   up:  fe80::ec4:7aff:fe48:69cb/64  1500 70.034 MB 65.839 MB
    br30 [        ]   up:      1500  0.001 MB  6.823 MB
    br30 [        ]   up:  fe80::212:c0ff:fe80:2676/64  1500  0.001 MB  6.823 MB
    eth0 [    br10]   up:                               1500  0.084 MB  0.689 MB
    eth1 [   bond0]   up:                               1500 71.769 MB 90.610 MB
    eth2 [   bond1]   up:                               1500  0.000 MB  4.296 MB
    eth3 [   bond1]   up:                               1500  0.001 MB  7.830 MB
    eth4 [   bond1]   up:                               1500  0.000 MB  4.296 MB
    eth5 [   bond1]   up:                               1500  0.000 MB  4.296 MB
      lo [        ]   up:                              65536  1.894 MB  1.894 MB
      lo [        ]   up:                      ::1/128 65536  1.894 MB  1.894 MB
   vnet0 [    br20]   up:                               1500 60.653 MB  0.136 MB
   vnet0 [    br20]   up:   fe80::fc54:ff:fec4:2984/64  1500 60.653 MB  0.136 MB

compare this to ifinfo

root@unison-mgr-1:~/net-tools# ifinfo 
device:	address/netmask			MTU	   Tx (MB)	   Rx (MB)
bond0:	addr not set/mask not set      	1500	    68.449	    86.434
bond1:	addr not set/mask not set      	1500	     0.001	    19.761
br10:       	1500	     0.080	     0.207
br20:    	1500	    66.794	    62.806
br30:     	1500	     0.001	     6.508
eth0:	addr not set/mask not set      	1500	     0.080	     0.657
eth1:	addr not set/mask not set      	1500	    68.449	    86.434
eth2:	addr not set/mask not set      	1500	     0.000	     4.097
eth3:	addr not set/mask not set      	1500	     0.001	     7.469
eth4:	addr not set/mask not set      	1500	     0.000	     4.097
eth5:	addr not set/mask not set      	1500	     0.000	     4.097
lo:            	65536	     1.807	     1.807
vnet0:	addr not set/mask not set      	1500	    57.862	     0.130

Note that its not obvious what interface is associated with what bond/bridge in ifinfo.

Still a few bugs I need to quash, but should be ready for general use soon.

Viewed 33214 times by 1375 viewers

Unison Ceph beats reference architecture, including the flavor with NVMe drives

The paper is here. We focused on our product mix and the rough comparables in the report. Our units are immediately available as well, preloaded/preconfigured with Ceph.

The takeaway is this:

Whats really interesting in this is that the 36+2 reference architecture makes use of 2x NVMe drives. And as you can see, they really don’t help much in the tests.

This is not to say NVMe is bad; its not. It is to say that overall architecture matters. You can’t build fast/efficient systems out of a box of software and random collections of hardware. You really do need to pay attention to design to get the best performance. You also can’t just drop a silver bullet plugin card/disk into a unit, and hope it does better. It won’t.

I should also point out that we used several generation older drives and SSDs (where we used SSDs) in these tests, while the reference architecture was kitted out with the latest and greatest. This is important, as an architecture that can’t drive systems efficiently won’t make effective use of its resournces, while an architecture that can use its components effectively will do often far better. We saw this 2 decades ago in the R8000 vs Dec Alpha days. There was no possibility that the R8000 could beat the Alpha on computational code … except .. that .. it did. Architecture matters. Far more than other aspects.

If your architecture can’t move the data, can’t process the data for storage, can’t write it out quickly, then chances are, you really shouldn’t be using it. Massive data storage requires very well architected systems, that can move data within and without.

This excellent performance coupled with the tremendous density (and therefore capacity) we bring to the table

means you can store your data, and work with it. This means you need to buy less kit to achieve your performance and capacity goals. Which reduces TCO and many other metrics.

As we like to say, with us, you’ll spend less and do more. With others, the other way around.

Viewed 38097 times by 1539 viewers

Nominate your favorite HPC product and company for a readers choice award

Please go here and nominate!

Last year, our customer Lucera, won best in Financial Services. We built the vast majority of their infrastructure, so we like to think we contributed in some manner to their success.

This year, please don’t hesitate to nominate us (or second/third/etc.) for Best HPC Storage Product of Technology for Scalable Informatics Unison product, or whatever you’d like.

In addition to the nomination for Unison in storage, I put in nominations for Cadence in Financial Services, and in Data Intensive computing.

Viewed 44098 times by 1654 viewers