Profile cover photo
Profile photo
Jens Axboe
Linux kernel IO guy
Linux kernel IO guy

Jens's posts

Next time the networking guys say storage steals ideas from them, I'll just sit quietly and point at xmit_more.

Dear Google - if someone sends me spam email (which is caught, and marked as spam) with a meeting invite, don't tentatively add this to my calendar. It's spam.

Dear lazyweb...

Recently migrated my mail off a personal server and to Google Apps. After some slow syncing of old mails, finally getting set. So far, so good.

But this means replacing procmail rules with gmail filters. And I can't seem to get the settings right. Basically I want mail from a list to be labeled appropriately. That part is easy enough. But lets say I'm CC'ed on an email and I get the email twice. I want the filter to catch the one from the mailing list (which is then labeled and "archived", the latter seems to be gmail speak for "removing the inbox label"), but I want the one that gets to me directly to show up in the INBOX. Pray tell, how is that done? Googling hasn't helped me, maybe I'm special...

Added a new fio feature today, that allows you to specify a max tolerated latency and fio will then run your workload and tell you what the max iops and bandwidth the device can support given that constraint. 

Nexus 5 is a great phone, love it. But Google Now so blatantly demonstrating that Google tracks your every move is pretty damn creepy. 

Lets say you have a logging format, with standard items like time, value, and a few auxiliary other values. You look millions of these, perhaps even billions, to the extent where run time memory overhead can become problematic. For most use cases, the auxiliary values are identical, or perhaps just repeat a lot. What would be a good way to reduce the memory overhead?

The logging side should be very fast, with most processing of the log format postponed until when the log is read and/or analyzed.

blk-mq is finally merged. Much rejoice was heard! Now to push the virtio-blk conversion with the second pull request in this merge window. 

blk-mq/core posted for review, lets hope we can squeeze it into the 3.13 kernel.

In worse news, my lobbying for an nvme based test device has been fruitless. I'd like to get my hands on an nvme device for testing purposes. If anyone has contacts that would make that possible, please do reach out to me (either here, or Thanks!

Weekend in Vegas done, now packing bags for the Kernel Summit in Edinburgh.

The block multiqueue branch has been rebased on 3.12-rc5 now, in preparation for posting for review. Shaohua converted the tagging to the per-cpu ida, which is a nice improvement for the tag starvation case. Looks like it's marginally slower for the non-contended case, however.
Wait while more posts are being loaded