[svlug] Silicon Valley economic woes (was: Linux and hardware)
J C Lawrence
claw at kanga.nu
Tue Aug 21 23:58:02 PDT 2001
On Tue, 21 Aug 2001 22:45:15 -0700
Deirdre Saoirse Moen <deirdre at deirdre.net> wrote:
> JC Lawrence wrote:
>> We're still looking for someone who has a decent level of clue
>> developer-wise about networking, who can intelligently sit down
>> with a packet trace, can design and build protocols that actually
>> work and have some concept of (E)BNFs and state machine
>> synchronising, who understands what an ethernet frame is, what
>> the first three packets in a TCP stream are, what AL5 means, and
>> all the other normal things that network level developers do.
>> Its going on months now....
> Yes, that's fine -- if you care about how a TCP/IP stack works.
Actually I'm not talking about understanding a TCP/IP stack, just
low-level networking. Still very much at the application level,
still using standard BSD-style socket calls, but able to design and
write code that does new things at that level. Its just above
> However, there are those who, like me, can't be paid enough to
> care about anything much below the applications layer, not even 7
> figures (one per burrito layer) would do it. Sorry, but for Linux,
> the only development work right now is:
> A) Kernel B) Device Driver C) TCP/IP stack
Add (that I know of):
G) Content cacheing/peering/distribution
H) Routing (MPLS, OSPF, BGP, et al)
L) Network management (cf/ala Tivoli)
(the latter two are particularly true if you add in the *BSDs, with
Solaris helping a fair bit too).
> Which is funny given that it's got A and C already working and a
> lot of B already. But I digress.
On the TCP/IP side, Linux does not make a good core router at this
point, it doesn't handle high sustained packet rates well
especially as a forwarding/packet processing engine (routing or
otherwise), there's relatively little work been done for Linux on
other protocols atop IP other than TCP (quite a bit of interest in
that area specifically), it doesn't handle very large numbers of
open sockets well etc etc etc.
The problem isn't that Linux doesn't have a TCP/IP stack, or that
it isn't any good, is that it is good enough to be attractive but
also not quite good enough. cf Worse is Better.
> Imho, 99% of these jobs are for companies who will fail RSN
Given a choice between working and not working, I prefer to be
working even if I know the company will likely not survive. Keeps
Reviewing the places I interviewed at:
Almost all were startups. 60% likely won't get financing. 20%
will get financing at terms that will kill them. The remaining
20% will live. Perhaps not well, and likely only to an
(unfriendly/desperate) acquisition but they'll live. Heck, the
startup I just left just got bought by Intel (I was doing SSL
and Apache code).
> And there are a lot of clueful people who code for Linux, etc.,
> who work at:
> D) Applications layer.
I interviewed at a number of those positions as well. There is unmet
need still out there. I believe several of the positions I
interviewed for are still open. Your mileage apparantly varies.
> What you may be seeing is simply people who haven't clearly
> identified their own niche for themselves. That's called
> EXPERIENCE. How is someone going to know if they like or don't
> like low-level coding unless they try it?
Possibly. I'm generally more interested in what a person can do
figuring that they'll look after their preferences for themselves.
>> I'm still getting a couple cold-calls a week trying to get me to
>> leave my current position (which they know about) to go work for
>> them. I'm not that brilliant or special. There is work out
>> there, but the bar is higher and they're looking for people with
>> significant clue.
> This is the second time you've implied in a post that I'm
There no implication or assertion there, and I apologise for any
accidental implication that there is. Re-reading my post I fail to
see such an implication.
> Fortunately, I happen to know that's not the case.
My intent was to indicate that the current cries of despair
stratifies cleanly against skill levels, and to comment usefully on
> The fact is that, given development cycles, there's periodic
> demands and the demands for applications-level developers isn't
> there in ANY sector I'd work in. There are times when device
> driver people aren't in high demand, etc. A few years ago, Unix
> kernel people couldn't get jobs if they paid people.
Its possible that I'm entirely unaware of the dynamics in upper end
app deve. I've presumed that the pattern I see for mid-level and
low-level app dev carried through into upper level. Given the
reported extreme pain in web dev, it could be that there is a strong
disjoint in the curve as mid-layer app dev moves into upper level.
I haven't seen it. I don't see sign of it. That doesn't mean its
> Right now, kernel people are in demand and not just for
> Linux. I've gotten several calls looking for BSD kernel people
Yup. Quite a lot of platform porting work going on out there, some
to quite odd/unusual architectures too (eg Linux on printers as the
printer host OS). There are also a companies developing new
hardware architectures (CPU and system) looking for porting help.
>> The shape of the work/clue graph has changed,
> No -- the shape of the TYPE of work available has changed.
Could be. I haven't noticed this and haven't attempted to graph it.
> For people like me, I got caught in a double-whammy -- I've
> typically changed fields to another type of application
> programming when one went dry. However, I've always done R&D
> rather than MIS programming (and it's a point of pride) and that
> area has been hardest hit.
Yup. I typically do low-level app dev, falling back on QA dev
(never QA work) when necessary or useful to learn/gain a new field
(eg I wrote a QA harness for Sun, learning about firewalls in the
process). I've never done what I think you call MIS work and have
no interest in it.
> Applications development for new startups (those currently looking
> for kernel/driver/stack people) is probably 12-18 months off.
Skill-wise I've generally sat somewhere between the middle of the
application space and low-level back-end work just above the kernel
(I've not spent much time in the kernel). I can't comment well on
the upper half of application development, especially tiered systems
and the like as its not my field. As I commented, I understand that
web-dev was hit hard (which I'd count as a tiered system dev). It
may be that other upper level app dev was hit harder as well, but if
so I haven't seen it.
Two of the positions I interviewed for were in the upper end of
the applications layer (both were too far out side my competancy)
Four of the positions I interviewed for were in the upper middle
end of applications dev (agents, distributed management, content
distribution). (Didn't like the company, personality conflict,
too slow to decide/respond (x2))
2Wire until recently had a Java/servelet (on Linux)
architect/developer position open and couldn't fill it for the
life of them (no candidates).
> The point, I suppose, is that I have enough clue to know what my
> niche is and know that it's dry and NOT pretend to be something
> I'm not for money. And there's a lot of us out there and you're
> not respecting that not everyone will know or care about the type
> of work your company, whatever it is, may have.
No. I merely expect you (generically) to change and adapt as the
field changes and adapts. A friend (graphics guy) has the savings
to wait out the market for a year or two until work that is
specifically interesting to him comes around again. A fair choice
if you're in that position.
In the general case I'd rather good people went to work for Nominum
or Ponte or Parnassus and any of a score and more of other companies
doing good and interesting things. My current employers are just
one, very narrow, very specialised, very limited need, instance of
such, and two engineers (current open req (the other one is for a
Mac dev guy) is hardly going to solve Valley job needs.
> Fortunately, as you said, work for sysadmins is starting to pick
> up, so there are other options, for which I'm thankful. I've done
> it before and enjoyed it.
Watching Dice, Monster, BrassRing, Guru etc I see little sign that
need for good SysAdms has fallen much if at all.
J C Lawrence )\._.,--....,'``.
---------(*) /, _.. \ _\ ;`._ ,.
claw at kanga.nu `._.-(,_..'--(,_..'`-.;.'
http://www.kanga.nu/~claw/ Oh Freddled Gruntbuggly
More information about the svlug