I managed to successfully install hardware and software.  Only drawback is that the ioFusion2 drive in some way conflicted with one of my NIC cards, so I had to move the ethernet back to the Intel controller.  The E1000 chipset drivers don’t work right with hardware offloading in Linux so it is necessary to disable it which puts more burdon on the CPU.  But I doubt it will make a measurable difference, between hosts I am seeing .1ms ping time.

     So drivers worked properly with the 5.10.20 kernel I am now running, and the utility software all installed and functioned properly, so the database is now operating off of it.  It reduced our WordPress home page load times from 240-290ms region to 207ms.  Database which used to max at 400TPS has successfully handled bursts of 2923 TPS.

     Database speed issues may or may not be resolved tomorrow with maintenance.

     The preferred way of using this drive is to invoke a feature called Atomic Writes.  What that does is it bypasses cache on a write and writes directly to flash memory eliminating the need for the database to do a double write while still insuring data integrity.

     The challenge is that drivers have not been officially supported for these drives since Sans Disk bought them and that was kernel release 3.6.x and we’re on 5.10.x.  However, a volunteer effort has brought the drivers forward for kernel 5.10.x but aren’t well tested.  And while I’ve located the drive code I still have not found the fusion I/O file system tools yet.

     Tomorrow the hardware issues will be addressed but the software may or may not.

     The Fusion I/O Flash drive that Amazon first told me would arrive on March 5th, then updated and said it will arrive between March 5th and March 12th, arrived today.

     I will be installing it on Friday evening.  During this time most services will be unavailable for probably about 1/2 hour starting around 11PM.  When completed, this should resolve the database bottleneck.

     I had looked at Oracle Linux as a possible replacement for CentOS down the road, however, Oracle is too broken.  First, x2go will not work because of an incomplete perl environment, and second, it is tied to 3.10 kernel which is not safe on modern Intel CPUs and lacks many features of modern kernels.  As a result we will not be offering an Oracle Linux shell server.

     Kernel upgrades are finally completed.  It took abnormally wrong because something went wrong with the kernel install on mail and it ended up with a grub configuration pointing to two ancient kernels that no longer exist and thus it was unable to boot.

     I was able to fix the grub configuration with boot repair but then had to manually mount all the shell servers NFS partitions from it as it was dead when they booted.

     I am planning yet another kernel upgrade tonight starting at 11pm as there is an issue with the kernel taking memory from the wrong end of a linked list resulting in poor guest performance.  Since we use VM’s to implement most of our services, guest performance is important.  If this goes as smoothly as the last upgrade it should be completed well before midnight.

     With respect to MariaDB database hardware, I’ve been notified by Amazon that it will be as much as a week later than originally promised, so the upgrade may not happen until the weekend of March 12th.

     I found one of MariaDB’s recommended drive model on Amazon for a very reasonable price and ordered.  I don’t know if my motherboard will be able to boot from it because I don’t know if the BIOS knows about it, but it should be compatible from a hardware standpoint and Linux talks directly bypassing the BIOS once booted and all this drive is going to be used for is MariaDB storage so it should be fine.

     So thank you to all of you who sent me suggestions and web references.  I expect to have it in hand next Friday and baring no snow will install it that weekend, else will wait for a weekend without snow.

     I recently turned up a Hubzilla in addition to the Friendica Fediverse social media site here.  Sunday it seemed to be okay but come Monday the server was seeing loads of around 150 or so at which point a number of web apps turned themselves off.  Also it was exhausted on connections which made many peoples WordPress sites break.

     After much fine tuning I’ve gotten MySQL to handle a touch over 400 transactions per second, but occasionally hubzilla spikes higher than that and the load will temporarily rise.

     After doing some research on MariaDB and tuning it as best I could, now load only goes up to about 8 as opposed to 150, but virtually all waiting processes are waiting on DIsk I/O at that point so it appears that peaks are saturating the drives I/O capability.  Already this is on a 4-disk RAID 10 array with everything pretty finely tuned.

     Researching I am finding most people are not getting much better than 400 TPS on rotary drives and perhaps 500 TPS with M.2 SSD drives, but with a PCIE SSD drive that supports atomic writes, up to 1500 TPS has been obtained.

     So I am searching for same but not having much luck.  I can find lots of M.2s but can not find PCIE drives.  When I search I only find PCIE SSD adapters, not what I need.

     Does anybody know where I can source these and who makes them, perhaps some make and model numbers?

     Wave cable uses a company called Synacor to provide their e-mail.  Synacor is allowing spammers to use their servers.

     We have had this problem ourselves in the past with hacked accounts being used to spam.  We have solved it by installing fail2ban to limit brute force password attacks and postfwd to force a password change if/when a clients account is abused.  It detects abuse by looking for a customer coming from IP’s in three or more countries in 24 hours.  This pattern only happens with botnets which spammers are using to send their spam.

     If anyone knows how to contact their network or mail administrator I’d be glad to point them at these resources.

     When spammers use unclean e-mail lists that contain addresses that don’t exist on our server, it triggers a program called fail2ban to ban them after several unsuccessful attempts because it thinks someone is probing for user addresses to spam.

     I have unbanned their servers twice but as long as they continue to allow their resources to be used for spamming, mail delivery from wavecable.com is going to be problematic.

     If you are a customer of wavecable, it may help to complain to them that your e-mail is not being delivered because of other customers on synacor.com’s servers abuse them.