1,407 reputation
1711
bio website redhardsupra.blogspot.com
location Laurel, MD
age 37
visits member for 5 years, 6 months
seen 2 hours ago

Feb
15
comment limit linux background flush (dirty pages)
What's your methodology to arrive at the '15MB for dirty_buffers is optimal' part?
Jan
20
comment 2.5 inch drives versus 3.5 inch drives
I could see the contrary: longer tracks contain more data, thus as long as the file system places clusters similar data, the need for seeking altogether might be lower, even though the duration of the seek itself might be longer. From some benchmarks I've seen, seeking just one track takes miliseconds, and it's not that much smaller than seeking a full inside-to-out distance. So I'd rather have denser media, or longer tracks (or whatever allows me to do a sequential read without seeking), than lots of 'fast' seeks. I guess what I'm trying to say is there is no better seek than no seek at all.
Dec
9
comment Setting MTU Per Destination Host
Dunno about OSX, but in normal *nixes, MTU tends to be set per interface, not per host. Maybe you could create a bunch of virtual interfaces with individual MTU's, and give them IP's on different networks, so routing would take care of picking the right interface, and the different MTU in the process? Just an idea... don't shoot me.
Dec
6
comment tuning linux cache settings for inode caching
Yup, I run updatedb from cron to keep the inode cache 'warm.' It works well, take a look at your slabtop to see statistics. Also, running strace -c updated can give you some insight as far as how much gets updated.
Dec
5
comment Encryption between a LAN switch and an end device
IPSec is a higher layler protocol, do you specifically want Link Layer encryption?
Dec
1
comment How do I install xen hypervisor + linux onto a usb memory stick
What have you tried so far? Most usb sticks just show up as a regular /dev/sdx, so you should be able to pick that as a drive and install onto it. If the installer does not want to work with a non-hdd, you can always make a ext{2,3,4} partition on the USB stick and make it boot with EXTLINUX, point it at the hypervisor, and let it boot.
Nov
28
comment TCP/IP & throughput between FreeNAS (BSD) server & other LAN machines
How fast is the read speed on the same file off the disk, without any network in the process? Lots of current-day hard drives have speed of about 70MB/sec for sequential reads, so maybe you're just bottlenecked on the disk?
Jul
21
comment Linux tmpfs having no effect on performance!
The kernel tends to be much smarter than your average sysadmin on determining what should be cached and what shouldn't. There's a whole lotta caching going on of stuff you don't really get to see from user's perspective. I tend to clear caches only for testing. For actual production, I'd leave it alone in 99% of cases. Or to quote Agent Smith: Never send a human to do a machine's job. ;)
Jun
28
comment How to mirror filesystems with millions of hardlinks?
Anything that works at a block level, not file-system level would probably be a huge improvement.
Jun
15
comment Why would you use IPv6 internally?
These are all popular myths, here's the bit more info: IPv6's multicasting is mandatory for basic functionality: for example to do a IPv4's ping broadcast equivallent you ping6 to FF02::1 for all regular nodes, and FF02::2 for all routers. IPv6's IPSec does not change ANYTHING from IPv4. You don't get any security for free. Still gotta configure all the modes, and deal with key distribution. IPv6's autoconfiguration is utter junk; by default it's as insecure as MAC<->IPv4, and it does NOT hand out DNS. If you want DNS you gotta install DHCPv6, so no gain there.
Jun
8
comment Why is a single drive much faster than 4 of them in RAID5?
gets the accept for guessing 'bad drive' correctly.
Jun
8
comment Why is a single drive much faster than 4 of them in RAID5?
straight out of lspci: LSI Logic / Symbios Logic 53c1030 PCI-X Fusion-MPT Dual Ultra320 SCSI and Kernel driver in use: mptspi. I have upgraded the firmware on the controller, I'll try to retest some of these cases, just to see if it makes any difference.
Jun
8
comment Why is a single drive much faster than 4 of them in RAID5?
I know it's bad, but I don't think it's 8MB/sec bad, is it? That's kinda useless in today's world, when you can grab a USB enclosure with a $50 drive and get 30mb/sec out of it.
May
30
comment Storing and backing up 10 million files on Linux
Yup, locality is crucial. Look at your usage patterns. Most problems tend to follow the Pareto Principle (80% of processes hitting 20% of data), so if you could figure out which files need to be cached in RAM, or just put on a separate partition with a different layout of directories, so it takes less directory lookups or seeks, it would probably help a lot. Spreading the frequently accessed files on different spindles of disks so seeks could be done in parallel could also help. +1 for @nemo for bringing up locality of reference.
May
30
comment Disk full, du tells different. How to further investigate?
The size of a file and the amount of space it takes on a filesystem are two separate things. The smaller the files tend to be, the bigger the discrepancy between them. If you write a script that sums up the sizes of files and compare it to the du -s of the same subtree, you're going to get a good idea if that's the case here.
May
26
comment Problem with Xen, xvda and sda
Why not mount them as xvda's then so it all agrees?
May
18
comment In tail -f, how do I filter out stuff that has certain keywords?
Yes, but they're merely symlinks to each other, not two different programs with some functional overlap. So wouldnt that be more of an 'addendum' (read: comment), not a full fledged answer? I've had downvotes for lesser offences...
May
17
comment different ACK behaviors (slowing down throughput?)
+1 for anyone that provides Dtrace code and output!
May
8
comment Rebuilding partition tables
Try PartedMagic's live iso, it has a mode where it can 'rescue' partitions by analyzing the disk's contents.
May
5
comment Is there a security concern exposing NTLM authentication over http or should it only be https?
'If you see it it's bad' is true. 'If you don't see it then it's good' is not necessarily true however. With challenge/response protocols you gotta study the protocol of what is being asked and how the response is generated, then seeing if you can obtain all the necessary information. en.wikipedia.org/wiki/NTLM#NTLMv2 has the details of what gets used. Time can be closely estimated, user and domain names can be determined, so the only truly random bit is 8byte random nonce, so that can probably be guessed in not too many tries.