baruch

11 ms·

61.
by baruch 6y ago
I've worked on a system that relied on it to do fast upgrade reboots and it worked. I'll have no problem relying on this again.
62.
by baruch 6y ago
A drive (be it HDD or SSD) is a block device, all drives are divided into blocks/sectors since they are a rather lossy medium. Data written to a bit may not necessarily be read accurately and so the solution is to have an Error-Correct
63.
by baruch 6y ago
Alibaba is about volume. AliExpress does single quantities too. Except for really cheap things and then you can several items together but nothing you'd really call bulk.
64.
by baruch 6y ago
He needs to do some more monte-carlo simulations on the jokes to get to the level of ToT. But it was a very nice presentation of the topic and a neat result.
65.
by baruch 7y ago
I got a pair from AliExpress. I didn't notice a significant change in noise cancelling performance (active or passive). But I'm not a picky listener anyway. YMMV.
66.
by baruch 7y ago
I've worked with plenty of HDDs and SSDs (software engineer for enterprise storage products, working on the hw-sw integration), HDDs and SSDs have pretty rapid failures. Sometimes you get several days or weeks of low-frequency errors o
67.
by baruch 7y ago
It's quite the opposite actually, a large company usually expected others to bend to its will and is not that interested in hearing opposed views and change their approach and that's what makes getting coffee into the kernel harde
68.
by baruch 7y ago
No. OS caches are not saved between reboots.
69.
by baruch 7y ago
The lane issue is why there is this push for GPUDirect, the bulk memory transfer goes through the nic that is closest to the GPU and straight to the GPU DRAM without the need to cross through the CPUs limited lanes. In my experience, this w
70.
by baruch 7y ago
At least for NVMe SSDs that is already solved, there is a standard that all NVMe SSDs implement and you only need one driver for all of them. If you want you can use SPDK or one of the few other drivers and you get the full speed block acce
71.
by baruch 8y ago
Intel bought one of the IB players and for all I know killed it.
72.
by baruch 8y ago
I've used it in a startup that did motivation data transfers and it was the only way around. It was and likely will is patented which makes it less attractive and the places where it is the only solution are niche.
73.
by baruch 8y ago
Yes, it works over udp. I've worked in the past in a startup that used fountain codes to transmit data over multicast and it worked beautifully. There was no need to retransmit specific packets and we only need to know when everyone go
74.
by baruch 8y ago
Nobody also peed on their wheels either. But they do need to handle quite a bit of dust and there is no one around to grease them. The APCs I had to use wouldn't survive that long without service.
75.
by baruch 8y ago
Did you know that your HDDs performance may drop if you have too many correctable ecc errors? We had servers that started an alert and a buzzer when ecc errors exceeded some internal threshold. The buzzer was making the chassis vibrate at t
76.
by baruch 8y ago
We are not yet up to Facebook power at weka.io but we fully use D and support some development of it. We have also released recently a library we use internally for user space threads and other support stuff, called mecca. D is really nice
77.
by baruch 8y ago
Not strictly a JIT but a preprocessor at compile time that makes sprintf calls faster by pre-splitting them: https://github.com/h2o/qrintf
78.
by baruch 8y ago
Disk sectors remained 512 for a very long time, only in recent years (2010 sounds right) did they make the 4k sectors with emulation for 512 or just native 4k drives. It is true however that sequential access to hdds and floppies was always
79.
by baruch 8y ago
I'm a fan of the user space threading for IO programs that handle more than one stream. It gives you the same context you are used to in blocking programs without the mostly needless thread context switching. My context is that most of
80.
by baruch 8y ago
At least in gcc you can use constructor attribute to run code before main. I've seen this used to initialize modules and create a dependency tree of modules to initialize before main even starts.
81.
by baruch 9y ago
NVMe and networking are now used from use space to reduce latency in storage systems. I'm doing that in my current job. There are libraries to help you with it such as DPDK and SPDK which are good starting points.
82.
by baruch 9y ago
There is a large overhead in SAS as it was designed for HDDs where the latency impact was not that critical. The next bottleneck for SSDs after the move to NVMe is the limited number of channels from a controller. NAND is pretty fast but it
83.
by baruch 9y ago
I've had a case that caused disk performance to drop because the memory correctable errors rose above a server BMC threshold. Turned out that the BMC beeped the internal speaker which vibrated the chassis at the right harmonic frequenc
84.
by baruch 9y ago
16.1 PB total writes guaranteed. About 0.8 DWPD from my calculation.
85.
by baruch 9y ago
There are quite a bit more complicated issues that require handling than just the erase count for each erase block. One part is that there are many things that the SSD controller does and keeps track of in order to increase the endurance to
86.
by baruch 9y ago
On one hand this is true, on the other, if you have two devices that can work for 5 years and then you anyway replace them because they are now too small and never failed you have one device that you replaced and the other hardly in use and
87.
by baruch 9y ago
Some of the meta data is only kept in ram and updated from time to time to special regions. These special regions can also suffer from endurance issues but are usually are SLC-use-MLC so they wear slower but it is a known failure mode for t
88.
by baruch 9y ago
It's not just the latest. Debian stable doesn't have the latest but it does have a larger selection of official packages.
89.
by baruch 9y ago
The one thing that will surely ruin your servers reliability is adding more ram to it. Based on my relatively recent experience a server with 1TB ram will have an MTBF of about 150k to 200k hours. Most of the failures (> 80%) are memory
90.
by baruch 9y ago
You either run it on a single CPU or the "os" layer needs to provide for thread migration. It also obviously needs to provide thread scheduling.

More ›