Hi. I'm jon.404, a Unix/Linux/Database/Openstack/Kubernetes Administrator, AWS/GCP/Azure Engineer, mathematics enthusiast, and amateur philosopher. This is where I rant about that which upsets me, laugh about that which amuses me, and jabber about that which holds my interest most: *nix.
Microsoft, why do you hate me?
I have a few ETL boxen that handle insane amounts of data every night. Hundreds of billions of rows of data get extracted from our OLTP databases, transformed, and then loaded into our datawarehousing machines. Needless to say, we're running into serious networking bottlenecks that are causing some ETL operations on larger databases to take 10+ hours, and each night we can watch the gigabit links on each of our ETL servers peg out at 100%. It's getting so bad that we have connections that get dropped (and so that given ETL operation would fail) and cause problems, so we decided to try some link aggregation (bonding, whatever you want to call it) to get more bandwidth to the ETL machines.
Ok. Every OS on the planet has some form of link aggregation capabilities...except for Microsoft Windows. Sadly, these machines *are* running Server 2003, so I looked into upgrading to Server 2008. Still not supported in Server 2008. Blarg. Someone please tell Microsoft that it's 2011 (link aggregation has been around for 15+ years now).
Alright, I discussed the issue with the reporting cluster administrator, and he's not comfortable with running Linux within the report cluster (which is mainly Windows based, with the exception of the actual databases, which are PostgreSQL on Linux). I mention the fact that Intel, nvidia, and HP NIC's can do link-level aggregation (as opposed to higher level "aggregation" in the kernel on sane OS's), so we look at the NIC's in the servers in question. Broad-effin'-com. Crap. Dell was kind enough to verify that the quad port intel cards they carry for their servers will indeed do link-level aggregation, and shipped me one for each server. At $500 per card, I'm sure glad we're sticking with Microsoft.
It never fails. Just when I think I can't stand dealing with Linux for another second, Microsoft goes and proves me wrong. I sure am glad I'm a Unix Admin and not a Windows Admin.
Ok. Every OS on the planet has some form of link aggregation capabilities...except for Microsoft Windows. Sadly, these machines *are* running Server 2003, so I looked into upgrading to Server 2008. Still not supported in Server 2008. Blarg. Someone please tell Microsoft that it's 2011 (link aggregation has been around for 15+ years now).
Alright, I discussed the issue with the reporting cluster administrator, and he's not comfortable with running Linux within the report cluster (which is mainly Windows based, with the exception of the actual databases, which are PostgreSQL on Linux). I mention the fact that Intel, nvidia, and HP NIC's can do link-level aggregation (as opposed to higher level "aggregation" in the kernel on sane OS's), so we look at the NIC's in the servers in question. Broad-effin'-com. Crap. Dell was kind enough to verify that the quad port intel cards they carry for their servers will indeed do link-level aggregation, and shipped me one for each server. At $500 per card, I'm sure glad we're sticking with Microsoft.
It never fails. Just when I think I can't stand dealing with Linux for another second, Microsoft goes and proves me wrong. I sure am glad I'm a Unix Admin and not a Windows Admin.
Total Comments 4
Comments
-
oh yea, they can overhead anything
Posted 09-17-2011 at 03:36 AM by Web31337 
-
The saga continues. Dell sent the wrong cards, so I ordered risers that would work with the cards. Dell doesn't make risers that work with PCI-X on the R410. Hrmmm.
Had them exchange the cards, but Dell doesn't make any risers that work with PCIe and R410's. Serious WTF? Hrmmm.
Grabbed an unused R610 so we could use the Intel cards, only to find out as we're nearing testing that Broadcom has released a link-aggregation driver for the built-in broadcom cards in the R410. At that point the only coherent phrase I could muster up was '#$@^*&$#!!!'
Blarg. Broadcom it is. At least now we don't have to purchase all new R610's + Intel NIC's and migrate everything over.
And to think...none of this would've happened if Microsoft supported link aggregation natively. Sigh.Posted 09-22-2011 at 11:00 AM by rocket357
-
Yeah - know that - been there.
Happy today that my own systems about 6 PC's and Laptops run on Linux today. Windows is where it should be in VM.
Remember the times I had to restart my Windows Servers minimum once a week to keep them running happily whil I restarted the UNIX machine once every 6 months. And the sole reason was to give the case a good cleaning and change filters.
Ever tried some creative way of aggregating the data? Comes to mind that more bandwith is usually quickly eaten up by more data. But it sounds as if you know your stuff.
As for hardware manufacturers - they increasingly start to act like MS and build systems but not to the end and leave you with unfinished solutions. Not thought through and always something new which won't fit into the old stuff.
Good luckPosted 06-28-2012 at 02:18 AM by error_401
-
I actually don't work for that company anymore, error_401, so unfortunately I can't say as to the long term effectiveness of the solution (Broadcom aggregation) we settled on.
Where I'm at now, I work solely on Linux and BSD, and virtually all of the NICs are paired into bonded pairs for all of our customer's critical HA servers. I work in Enterprise, so everything is redundant and scalable (well, in theory, at least haha). My previous company would've compared to what we call an "Agency" customer (tons of small sites with little failover capability)...so opportunities to solve silly problems like the one in this blog post are few and far between at my current job. Now I'm working more on architecting solutions and building infrastructures than trying to figure out why connections get dropped each night (though that does happen here on occasion, too =).
Perhaps I'll contact my buddies at my previous company and ask how the Broadcom NIC bonding worked out...Posted 07-04-2012 at 03:10 PM by rocket357
