Maximum 100 Mbps ethernet connection....
We should be using fiber, which also costs less power and is orders of magnitude faster.
If the words "enterprise-class" actually mean something more than "much larger markup than purchasing components" then go with something like http://www.marvell.com/company/news/pressDetail.do?releaseID=3576
For example, maybe the http://www.mitac.com/business/gfx_servers.html people have benchmarks representative of our DB/cache usage patterns. It's not like we have anything special (or x86-specific, Jasper!) other than very high bandwidth.
At least put out an RFP, please.
"It's not like we have anything special (or x86-specific, Jasper!) other than very high bandwidth."
Wikimedia Labs uses x86 hardware virtualization (just one example). We already have transit linkages that include fiber, and new fiber is far from cheap.
You persist in ignoring the costs of buying equipment. In terms of orders of magnitude, ~500 servers * $200 per server = $100,000 already. That is a conservative $200/server estimate that also doesn't take into account labor and other costs. To the level we'd want it, it's at least one more order of magnitude more expensive (no, we're not going to recover the costs by selling our existing servers).
On Sun, Dec 29, 2013 at 10:10 PM, James Salsman jsalsman@gmail.com wrote:
Maximum 100 Mbps ethernet connection....
We should be using fiber, which also costs less power and is orders of magnitude faster.
If the words "enterprise-class" actually mean something more than "much larger markup than purchasing components" then go with something like http://www.marvell.com/company/news/pressDetail.do?releaseID=3576
For example, maybe the http://www.mitac.com/business/gfx_servers.html people have benchmarks representative of our DB/cache usage patterns. It's not like we have anything special (or x86-specific, Jasper!) other than very high bandwidth.
At least put out an RFP, please.
Wikimedia-l mailing list Wikimedia-l@lists.wikimedia.org Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, mailto:wikimedia-l-request@lists.wikimedia.org?subject=unsubscribe
Oh man ---
The trolling here is amazing!!!
First off, you can't just plug a fiber cable into an ethernet socket. You need at least SFP+'s for 10G fiber connections, plus the cost of fiber itself -- it's way more expensive than a 100mbit connection.
That said, we do use SFP+ based 10G for our varnish layer -- the amount of traffic pushed by those boxes compared to the amount it would cost to buy a bunch of additional machines makes the extra cost of the interfaces (and don't forget the switch ports! 10G switches cost more than 100mbit switches) make perfect financial sense.
Leslie
PS - someone technical may point out that DAC (which is copper! gasp!) is cheaper than two SFP+'s and a fiber cable. That's true! However we have had major issues with DAC compatibility on the switch side, and some issues on the hardware side. SFP+'s rarely (albeit occasionally) have that issue. Plus we can keep a spare inventory that is usable in multiple places, making our sparing easier and life easier on the DC techs when we need to scramble for an unexpected 10G need.
On Sun, Dec 29, 2013 at 10:10 PM, James Salsman jsalsman@gmail.com wrote:
Maximum 100 Mbps ethernet connection....
We should be using fiber, which also costs less power and is orders of magnitude faster.
If the words "enterprise-class" actually mean something more than "much larger markup than purchasing components" then go with something like http://www.marvell.com/company/news/pressDetail.do?releaseID=3576
For example, maybe the http://www.mitac.com/business/gfx_servers.html people have benchmarks representative of our DB/cache usage patterns. It's not like we have anything special (or x86-specific, Jasper!) other than very high bandwidth.
At least put out an RFP, please.
Wikimedia-l mailing list Wikimedia-l@lists.wikimedia.org Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, mailto:wikimedia-l-request@lists.wikimedia.org?subject=unsubscribe
As a non-technical person on this list, I have no idea what you just said Lesile (and Tim, earlier in this thread) but I greatly enjoyed reading it! Keep up the good work :-) -Liam / Wittylama
On Tuesday, 31 December 2013, Leslie Carr wrote:
Oh man ---
The trolling here is amazing!!!
First off, you can't just plug a fiber cable into an ethernet socket. You need at least SFP+'s for 10G fiber connections, plus the cost of fiber itself -- it's way more expensive than a 100mbit connection.
That said, we do use SFP+ based 10G for our varnish layer -- the amount of traffic pushed by those boxes compared to the amount it would cost to buy a bunch of additional machines makes the extra cost of the interfaces (and don't forget the switch ports! 10G switches cost more than 100mbit switches) make perfect financial sense.
Leslie
PS - someone technical may point out that DAC (which is copper! gasp!) is cheaper than two SFP+'s and a fiber cable. That's true! However we have had major issues with DAC compatibility on the switch side, and some issues on the hardware side. SFP+'s rarely (albeit occasionally) have that issue. Plus we can keep a spare inventory that is usable in multiple places, making our sparing easier and life easier on the DC techs when we need to scramble for an unexpected 10G need.
On Sun, Dec 29, 2013 at 10:10 PM, James Salsman <jsalsman@gmail.comjavascript:;> wrote:
Maximum 100 Mbps ethernet connection....
We should be using fiber, which also costs less power and is orders of magnitude faster.
If the words "enterprise-class" actually mean something more than "much larger markup than purchasing components" then go with something like http://www.marvell.com/company/news/pressDetail.do?releaseID=3576
For example, maybe the http://www.mitac.com/business/gfx_servers.html people have benchmarks representative of our DB/cache usage patterns. It's not like we have anything special (or x86-specific, Jasper!) other than very high bandwidth.
At least put out an RFP, please.
Wikimedia-l mailing list Wikimedia-l@lists.wikimedia.org javascript:; Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l,
<mailto:wikimedia-l-request@lists.wikimedia.org javascript:; ?subject=unsubscribe>
-- Leslie Carr Wikimedia Foundation AS 14907, 43821 http://as14907.peeringdb.com/
Wikimedia-l mailing list Wikimedia-l@lists.wikimedia.org javascript:; Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, <mailto:wikimedia-l-request@lists.wikimedia.org javascript:; ?subject=unsubscribe>
wikimedia-l@lists.wikimedia.org