Using a CacheFlow Enterprise Cache from 2001

[@ZdenalAsdf]

The two hardest things in programming are, in fact, naming things, cache invalidation, and off-by-one errors.

[@klaernie]

I definitely cite this more than once a month.

[@doodlebroSH]

gottem

[@karan_hiremath]

I thought it was the 10 hardest things in programming

[@Momi_V]

​@@karan_hiremath Binary isn't that bad, at least compared to the other stuff...

[@ChasOwens]

threads Don’t forget.

[@KJ7BZC]

Even if the cache is slower or equal to the server itself, it can still help if you're serving dynamic content with something like ASP or PHP. Have the cache handle all the static content, prioritize your server processing power for the content that needs to be generated every time.

[@clabretro]

yes! I'll be doing tests against a DB-backed app in future cache videos

[@serpent77]

Was just coming to the comments to point this out :)

[@usagold8]

Yup. Would love to see how it handles dynamic content, especially cookie sessions 😜

[@Ziraya0]

Presumably you could also set up your router to loadbalance across multiple cache units, and one would hope the cache units are cheaper than the server. Letting you scale the cache for static content until the server is maxed out handling dynamic content

[@James-z6x3u]

Yes, I do this all the time. I'll cache dynamic content for a period to reduce the unnecessary database, HHTTP, and IO requests. You can also see this today being done to convert wordpress content to a static JAMStack site, while still using wordpress (a headless version) as the engine.

[@JMassengill]

Really enjoy your content. You don’t just sit and talk, talk, talk you get in the nuts and bolts of retro enterprise computing. Great video. Well done!

[@clabretro]

Thank you, I appreciate that!

[@ExtremeMetal]

He does manage to make even the talk, talk, talking interesting

[@carmelweston1041]

Nah, notice how clean that cache server was, wow. Sadly though, he's a toucher, soon as the top is off he has to touch PCBs, 😮 ah well. Well done getting it going, thanks for publishing.

[@uiopuiop3472]

Retro networking and server hardware is so much more interesting than retro gaming that most others do, loving this content

[@Aruneh]

Love it when you run period software on the servers. That ancient apache is just *chefs kiss*

[@tcpnetworks]

We had a thousand of these fuckers running as a cache for a "large Australian ISP" . Every state had hundreds, serving up "them Internets" to consumers on dial-up and low-speed DSL... The Internet was a vastly different place back then. Our largest links to the POPs were 50-100Mb/sec, for the whole state. SSL killed this stuff off. There was no way to cache site assets on disk for replay once SSL became the default. We scrapped all of it in 2004, just before Asia Online went broke, as we moved away from dial-up to DSL/Broadband. Hateful things.. :P

[@TomAtkinson]

As transparent proxy? I wonder what the utilisation was? Enough to make the machine pay for itself in how long? 9 months? I never touched but I heard we used transparent proxy at IHUG, but probably linux based. 1998-2000

[@tcpnetworks]

@@TomAtkinson We had racks of them doing proxy, and a pile of them doing the various websites. We are talking 24-25 years ago... They were good at saving bandwidth - but SSL killed it..

[@GeorgeTsiros]

hateful or hated? 🤔

[@casperghst42]

And today zScaler will do the same with https traffic where it repack the communication with it's own certificate chain.

[@hariranormal5584]

Telstra best

[@muchosa1]

I worked for a web hosting company from 1999-2015. I saw a lot tech come through the farms. Great time to be involved in the early internet and watch it evolve.

[@isharted]

I've managed Blue Coat (Symantec/Broadcom) web proxies for the past 15 years, and this is the first time I've seen its predecessor. The management CLI is still exactly the same ("enter" 3 times, install multiple systems to boot into to make upgrades and backouts easy, syntax similar to Cisco routers/firewalls). Even the web interface has a similar basic structure, and it took them way too long to move off of Java. As others have mentioned, configuring WCCP on your Cisco router/firewall was a popular way to handle routing at the time. PBR can work but it gets messy. Explicit proxy config or a PAC file would be an the option to have the client make the routing decisions. Great video!

[@1anwrang13r]

To paraphrase an old saying "Some people, when faced with a networking problem, think 'I know - I'll use Policy Based Routing!' They now have two problems"

[@theserialport]

Awesome cache work and great to see it working! Looking forward to seeing if the CacheFlow or another cache can be utilized in your network for client-side caching.. since fractional T1 is kinda slow, ya know

[@serpent77]

Woot! Yet another channel I've been binging lately :)

[@jfbeam]

I have a Cisco NM-CE (cache engine) in my collection. I never used it, did image the drive. It appears to have been part of Sprint's network (cellular???) caching a variety of things. (not surprised... a load of music files.) It'd take months of digging through the wayback machine to even find docs on how to use it. Cisco has long since deleted any reference to the code it runs. (I'd just turn it into a little linux server, but I have TV remotes more powerful than this thing. And I have a running UCS-E EHWIC in my router...)

[@423tech]

Hannah Montana Linux, I was NOT prepared 😂😂😂

[@_vilepenguin]

The fact you shared my glee with the Compaq mouse is one of the reasons I watch.

[@beardedgaming1337]

when i worked in IT at a bigbox repair center. we built a linux cache download server. god that changed our lives.... back then internet was maybe 20meg? at best. so being able to pull at 100 meg over the network was a game changer

[@tomorrow6]

@tripplefives1402back in 2001, cheaper 32 bit OS’s didn’t have very good large RAM support , and people tended to skimp on SCSI controllers cache and short stroking physical disks , so a cache could add quite a good bit of performance to a “typical” non optimized server

[@Daniel15au]

​​@tripplefives1402These days, separate server cache machines are still used, but more for geographic distribution as part of a CDN. Having a bunch of caches close to most of your users helps improve the performance of your web site/app quite a bit.

[@wirenoises]

This is a great trip down memory lane! Did a ton of large enterprise CF and BlueCoat work as a reseller-engineer. The products really did work well for that era of the internet. The CF/BlueCoat engineers were passionate for the tech and were always eager to engage finding solutions for complex and innovative implementations.

[@clabretro]

that's cool!

[@ivanmaglica264]

Gig of ram in 2001 was sort of like 1TB now. We had 128MB workstation and was still 4x more than average business PC had.

[@eDoc2020]

A low-end laptop from early 2000 would have 64 megs of RAM. A normal desktop these days is more like 16 gigs so I'd say 1GB in 2000 is more like 256GB today. 256 gigs on a server today is pretty pedestrian but we also tend to run more services per box so it's not quite a fair comparison.

[@mykolapliashechnykov8701]

RAM got cheap about 1998-1999. I remember seeing a SSD unit at Paypal's data center, MFG date 1998. 19.2Gb of SDRAM, SCSI and Fibre Channel connections, size of a decent UPS, lead-acid battery backup. According to the attached logs, it broke in 1999 and was never fixed. Most likely PCB corrosion from the venting battery. The manufacturer was some obscure company from Silicon Valley, no relevant data on the internet. Asked for some 256M ECC sticks for my Pentium 2 historical build, wasn't successful, bought 512M ones on Ebay.

[@TylerStartz]

My fascination with obscure hardware like this is immeasurable. Thank you for all the effort you put into these! Also, a fellow Andrew Camarata and Diesel Creek fan?? Right on!

[@cocusar]

this is a full course meal of 00's goodies. I have a dell latitude c640 and c610, I love those laptops. your video is amazing man, I always enjoy them.

[@questionablecommands9423]

3:46 ...is this bait for engagement? Well, you got me because the two hardest things in software engineering is naming things, cache invalidation, and off by one errors.

[@clabretro]

sounds like I was off by one item in my list

[@questionablecommands9423]

@@clabretro touche. It must've been in the list just before that time you found a nickel.

[@c1ph3rpunk]

Sitting at Yahoo and had to force a cache refresh on the letters M & S, poked the purge and watched racks of Sun E220’s all slam to 100% in like 2 seconds. Naming standards are easy, make sure you have at least 25 of them. 😂

[@dataolle]

The best thing about standards is that there are so many standards to choose from.

[@JoshuaDellinger-jo9nx]

I worked for CF back in 2001, and we had used Foundry Server Iron switches, and commodity desktop machines as the servers to run this exact test in the test lab. We had some customized SSL handshake cards in the chassis as well, because that was CPU intensive back then. So you could take that load off your web server and put it on the CF box. Amazing to see you recreate a part of computing history.

[@geoffjohnston2413]

Another great video! Of course you are a Diesel Creek fan, that was a nice shoutout to another quality creator.

[@seshpenguin]

The stacks (literally) are getting more involved and it's so much fun to watch!

[@nateguiger2641]

Very entertaining, your pacing and cadence are top tier.

[@AlexKidd4Fun]

Im an ancient Compaq/HP admin and remember these servers and even older. Proliant 800, 1600, 1850R just to nane a few. The early Compaq days used something called EISA utilities versus what the clone systems called BIOS. This is why you needed SmartStart utilities. I believe EISA was used by IBM as well. EISA was shockingly similar to how UEFI on modern systems work just without the encryption/security bits. It was way more flexible at the time than BIOS but less intuitive. Funny how things go in cycles eh?

[@clabretro]

cycles indeed. I actually hopped into the eisa config screen (not sure what to call it, but the BIOS-looking screen) but forgot to film it

[@Lambykin]

There was also at least one motherboard manufacturer back in the day (Gigabyte, I believe, using a 486 CPU) that also had EISA slots. It had an AMI Bios for the very basics, and then had a bootable EISA config floppy for the EISA side of things. Very expensive at the time, but it was a more economical solution than purchasing a name-brand server such as Compaq. As for IBM, I never encountered EISA in the equipment we deployed. The IBM stuff we used had proprietary architecture called Micro Channel - shortened to MCA. All ancient history now.

[@TrolleyMC]

That 5 dollar subscription to your patreon is so worth it, content like this is so hard to find.

[@clabretro]

It means a lot to hear that, thank you!

[@nzspambot]

Your content is starting to give me PTSD from my early IT days 😅 You should see if you can find one of the OG google boxes. We had a couple as coffee tables in my last job

[@rnts08]

I get both ptsd and nostalgia vibes, it's amazing. 😂

[@pauldunecat]

We had one in the office around y2k. I think it was a 1U yellow pizza box. 🙂

[@RomainBourdy]

@@pauldunecat I remember it as blue :(

[@rabidbigdog]

And yes, those Compaq Proliants and their heritage are still the best x86 servers. Really well made, great setup software and excellent performance. There are still bits of Digital/Compaq heritage in them today.

[@spewp]

I really liked the aesthetic of this era of gear. Reminds me of a neon yellow motorola pager.

[@firenado4295]

never thought I would hear diesel creek mentioned on this side of the internet, but then I'm here and I watch the both of you so it not that odd i guess.

[@ram89572]

You can be both a tech geek and a machines/equipment geek. You probably enjoy watching things work and fixing broken things, but don't get to actually work on all that stuff in your day to day life

[@firenado4295]

@@ram89572 I know but its still funny to me that I watch about 10~ channels on this website out of millions and one of those people watches a couple of the same channels I do in the same obscure overlap. I know its not too much of a jump from computer tech to mechanical tech cuz its all the same sort of trouble shooting at the end of the day.

[@edfromnc7660]

I worked for a very large hospital system and the enterprise people bought all of us Cache Flows (we had a pair). Our network went from our local hospital to national gateways to access the internet. We were using the Cache Flow to keep redundant traffic off the DS-3 to our gateways. I never could find direct evidence of how much it helped, but they were promoted as a big reduction in traffic.

[@clabretro]

interesting!

[@donwald3436]

wait you mean you ran them as forward (reverse-reverse) proxies?????

[@VK2FVAX]

I wish to gently correct you. The quote is: "There are only 2 hard things in computer science: (0) Cache Invalidation. (1) Naming Things. (7) Asynchronous Callbacks (2) Off-by-one Errors."

[@robertclark8351]

People used to hit each other with Teardrop for fun. That would peg a single-core Windows system so hard it would become unresponsive.

[@manitoba-op4jx]

back when everyone hooked their computer straight to the modem.....

[@1MountainLife]

I stumbled here from my recommendations which is interesting since I don't watch this type of content but you hooked me. Just subbed.

[@clabretro]

thanks!

[@justine1816]

42 minutes! Hell yeah. Love your longer vids.

[@clabretro]

awesome! I like longer ones too haha

[@miked4377]

great stuff clab! its really cool to see what you can do with this old tech!

[@grapowski]

So cool that you watch Diesel Creek too! Cheers from Brazil!

[@EdgarGrefve]

that yellow and black front panel look so cool they sure got the design right

[@un_simp1127]

Just about to sleep but new clabretro video so human neccesities can wait

[@IkanGelamaKuning]

Im a retired it tech. Retired at the end if 2021. Can no longer catch up new things. Getting old really sucks. 😂. However I enjoy watching your videos about old IT tech. Reminds me during my golden years.

[@donwald3436]

I'm a decade and a bit behind you but "can no longer catch up" is already hitting hard lol.

[@IkanGelamaKuning]

@donwald3436 🤣

[@vote4carp]

"Should we attempt to compile and run the Jolt2 DOS attack from the Hannah Montana Linux machine? Yes... the answer is obviously, yes." Amazing. Love it.

[@abysspegasusgaming]

Having bought an HP ProLiant DL380 G7 (2U rackmount), it's a pain and fun to tinker around with, at least no CD/DVD ROMs are needed to access any special features. Gotta love having everything built in.

[@--BiZ--]

This is the channel I needed in my life... and I didn't even know it..

[@doskungen]

Hannah Montana Linux.... I love your videos SO much Clabretro! Wish I had the space to do stuff like this myself! :)

[@trekker6000]

Huge Andrew Camarata fan here, love your channel!

[@MetaversaicCat]

This is one cache you might not want to clear. Gotta love the “System Busy” message. We have late 90s Point of Sale at work (Toshiba/IBM) with text terminal interfaces. The error “B000 SYSTEM BUSY” message is way too common.

[@sprint955st]

Your POS system sounds like IBM 4680. I supported those for a defunct British High Street chain during the 90s, and helped roll them out and replace ancient 3650 terminals and controllers.

[@MetaversaicCat]

@@sprint955st sounds about right, we have a 4690 system, that oddly enough boots over the network, and takes forever to startup. The main issue is the mix of old and new hardware just sometimes having a fit. I wouldn’t be surprised if they were using a pentium or some 80xx series chip. The slowness is real.

[@cambridgeport90]

I never knew that there were dedicated cache appliances back in the day till I watched this. Dang, but I love your videos. You have all the things I was around for, but my paths didn't cross in enterprise tech until about ten years after this was possibly out of date.

[@clabretro]

Glad you like the videos!

[@Nate-hf8hm]

Honestly I loved this video, Just an awesome piece of obscure tech

[@ScarlettStunningSpace]

Thank the KZbin recommendations

[@kidkool27]

Tail latency is one of the core metrics we track on our services, a drop from 3s to 100ms would be insanely good.

[@piketfencecartel]

When it was doing nothing I thought it might be looking for its host in a loop. I love the suspense, 9:30 into the video.

[@rabidbigdog]

Superb. The cache devices like this and software like Squid, Microsoft Proxy Server were actually REALLY effective. Mostly because the response time was the local LAN for the office for some material, not all piped out to the WAN/internet. This make a big difference to the small-pipe performance for office environments. The synthetic setup there could be retested with an artificial 2Mbps link to the end-point-web server to simulate an office T1-link, ISDN or even dial-up (!) internet.

[@kc0eks]

Enjoying this stuff so much. Thanks for sharing

[@c1ph3rpunk]

I’d build a CDN with it, a clabretro desk network. I forgot how much I liked Network World back then, it was one of the best. Ok, been meaning to, love your stuff, Fiber Channel Patreon membership earned.

[@easyy376]

We used something similar in the Army back up until mid 2010s and earlier. This wasn’t server facing but rather user facing. Before the days of starlink, we had like 5 Mbps satalite links. It was called a WAN scaler and it cached data so it would lower our network usage.

[@xtlmeth]

I watched the episode about the Nortel BPS 2000 switches. I used a bunch of those in the early 2000s. My favorite feature about those switches is when the activity LEDs blink, they blink for every packet that hits the port instead of the blinking on and off slowly on mondern switches. You could take a look at the switch and tell which ports were really busy and if you had a loop all of the lights would be solid.

[@SproutyPottedPlant]

That CacheFlow looks beautiful, like it belongs in a 90s music studio rather than an IT rack! If it was a ROMpler I bet it would have great synth pads 😅

[@nickwallette6201]

LOL - it really does! I have a StarTech 25U rack with the front half loaded with Cisco and Juniper gear, and the back half loaded with Korg, Roland, Yamaha, Emu, and Ensoniq gear. I guess this one would go right in the middle? haha

[@jcsalgueiro]

Instealled one of these, don't remember if this model, on a company once as internet cache. it had it's own internet connection and all users's requests were intercepted by the site router and were sent to Cacheflow using WCCP protocol. Worked like a charm and utilization on the company main 2Mbit international frame-relay network dropped a lot.

[@matthawaii]

Great video! Would be curious to see the results if you pumped up the simultaneous users from 5 to 100+. I imagine that is when a device like this starts to shine.

[@clabretro]

yes! I'll be doing more types of tests in future cache videos

[@pauldunecat]

BlueCoat is still around, they pushed heavy into transparent web proxies and that became their main business, then they were bought by Symantec. As for sending traffic, ya ever hear of WCCP; the Web Cache Communications Protocol? That Cisco gear you have, specially the Layer3 switches or the 7200 VXR, support it. 😀

[@LordChariot]

...and now Broadcom.

[@pauldunecat]

@@LordChariot Oh, they dead then. LMAO

[@LordChariot]

@@pauldunecat pretty much

[@ChefRex]

I was having a pretty bland day and then Clabretro uploads! 😄

[@nrgonline]

So good! Haven’t seen a Cisco pix since my days running a cyber cafe in 2001

[@АндрейМилованов-у9у]

if you have limited Internet channel or trafic (as it was in some regions and countries in 2001) this cache can help adopt work to those conditions. Sites had another sructure and caches like this effectively optimized trafic.Our company in 1995-2000-... used special cache computer under FreeBSD (Pentium on Intel chipset, later in 2002 or 2003 it was upgraded on Pentium III Tualatin) and SCSI disks on Adaptec RAID ( AHA-2940, later Adaptec SCSI RAID 5400S) for similar optimization because it was cheaper and more effective than brand solutions.

[@Crazyman23]

At the first few seconds you show a monitor with redhat Linux. That was the first linix I ever used. We had what was called the milk crate. It was 6 or 7 diffrent Dell computers put together to make one functioning pc. We ran redhat first then switched to ubobtu around 07 or 08.

[@HyenaEmpyema]

I love this, thanks. Teenage me was buying 10k Cheetahs on ebay back then, could barely afford them (and the scsi cards to use them). It was my version of hot-rodding. Since you're a masochist with old heavy gear) maybe do a video on fixed-frequency monitors. I used to buy those and solder them in to VGA connectors and play with the internal scan controls to get it to show a picture. Remember how expensive large screens were back then? Then by the 2000s you could buy after-market video cards that had special adaptors to drive them on PCs.

[@colinmoller4321]

There really is something so incredibly satisfying about getting to play with all the enterprise big-iron gear that we dreamed about even getting a glance at in the 90s... I almost count myself lucky that my office/workshop/storage is way too small to entertain more than a handful of retro machines, so I avoid the late-night eBay gremlin :D

[@onelastpicandillbegone]

while i was watching your video my phone went off with a test alert from my country but still a great video

[@Lambykin]

Those Proliant servers will outlive us all. Very well built. Off-topic, but you can quite likely revive that CD-ROM if you're willing to spare a few minutes on it. Remove the drive, and remove the cover, exposing the moving parts inside. Blow out the years of dust, and clean the optical head. It likely has a layer of dust on it that won't blow off, but may not be readily seen by the naked eye. Next step - lightly lubricate the moving parts with a light machine oil (sewing machine oil is good), or a silicone-based lubricant often found in spray cans. The old lube was likely a light grease that would have hardened a bit by now, and will be slowing the movement of the head. If the head assembly rides on rails, you will want to lightly lube the rails with the spray. Basically, lightly lube any moving parts of the assembly. CD-ROM drives in servers are rarely used, and they often get dirty over time. They don't usually 'fail' in a catastrophic sense, but don't operate properly due to years of dust build-up. For best control of the lubrication process if using a spray in a can, spray the lubricant into a shot glass, and apply this using a q-tip. Don't use WD-40 - it will evaporate in short order and you'll end up with the same problem again.

[@jscheunemann]

17:30 love to hear the shout out for Free Geek. Now I have to go back and see what new stuff I can buy since I live nearby them……

[@Ceseuron]

5:55 Most Seagate disks have their capacity noted in the model number. Those are 18GB disks, as noted by the "18" in ST318404LC. Cheetah disks, if I recall correctly, were 10K RPM and these are probably Ultra 160 SCSI.

[@tankgrrl]

Woohoo! Just starting this video and I'm already "this is gonna be nuts" [in a good way] 😁

[@MARUKU]

Not the dell laptop being rebranded as "mily cyrus" girl i'm in tears wtf great video though thx

[@mstandish]

I do not miss setting up and managing a squid cache.

[@please_eject_the_disk]

Great stuff. Would love to see you take a crack at a pair of late-2000s Riverbed Steelhead (EX?) units for file/traffic caching. We used to use those on either side of a T1-based intranet between PA and CA to speed up people's ability to use RDP or viewing/caching PDFs and other workplace files. Interesting technology that worked surprisingly well for what it did. They were essentially rebadged OEM Dell PowerEdges, I believe.

[@clabretro]

Interesting! I'll keep an eye out for those

[@EdwardJamesBickels]

I've got an old Riverbed server that was just a rebadged R210 II. It now lives as my pfSense router.

[@churchers]

Go for the og web appliance server and get a cobalt raq

[@BigBenAdv]

Ah.. The good old WAN accelerators. Those are a different beast though - they don't actually so much cache, as provide a local connection termination endpoint. Basically, each support type of connection to be accelerated will terminate on the unit via LAN - the accelerator talks to the client/ server on either end directly to broker the connection like a MitM. The accelerator then deals with all the handshakes and passes only the 'interesting' traffic (i.e. the actual DATA) over the WAN. Some may have additional compression/ deduplication etc. applied when doing so. Hence, it reduces the amount of traffic over the WAN and further reduces the 'waiting time' on handshakes. These days, high-speed broadband is so ubiquitous that WAN acceleration isn't that important and in some cases, wouldn't even work - e.g. when protocols like HSTS or similar are used to mitigate MitM, it's not very compatible. Probably the more interesting part of WAN acceleration would be the ability to 'bond' multiple connections by using proprietary protocols that allow out-of-order packet receipt and sequence re-assembly. This is still used in modern SDWAN, though not all purported 'SDWAN' products support this.

[@please_eject_the_disk]

@@BigBenAdv it was so many years ago now, but I'm almost certain that particular types of files, like PDFs, had some level of local cache at the unit, so that if you opened a file that hasn't been modified from a remote file server, it just used the local version. This being in addition to the WAN acceleration for other connections (it made a local ERP system usable; it used to take a few minutes to start up the fat client, but the accelerator brought it down to a few seconds). The transparent caching could have been an enterprise feature, an add-on, or something unique to the generation of the hardware. Foggy memory, but we did eventually ditch them for just site-to-site VPN over broadband because the speed was much better compared to MPLS over T1.

[@slightlyevolved]

28:25, the rule of IT, to fix the thing, you first need to fix the thing preventing you from fixing the thing.

[@StephenFasciani]

I like this stuff because it reminds me of early thousands Internet hosting and as a person born in the late 90s, I didn’t get to experience what it was like to be an adult in the 90s and in the early thousands and I did look up to someone when I was a kid Who did this kind of server administration/web hosting stuff I am now a cloud engineer, and I feel like with the advancements that software has made were missing out on a key component of hardware appliances that while can be solved in software the physical nature of having yet another appliance is intriguing to me

[@digitalsparky]

Another thing that you may not have taken into account - you can deploy the cache to edge datacenters, so not only does it save main server load, but it can significantly reduce time to first byte and general site latency. (I'm a network engineer and work in web tech).

[@H3adcrash]

A note on the DL380 and IDE drives.. I've used the optical drive connector for hard drives in my ProLiant 3000 and 5500 in the past. Unless they changed stuff up a lot between them and the 380, I'm pretty sure that would still work!

[@seannalexander]

The cache invalidation quote was somebody who worked at netscape and passed away in 1997. They did not understand what the world is now.

[@BlakeHelms]

The two hardest problems in computer science are: 0. Naming Things 1. Cache Invalidation 2. Off by One Errors

[@idahofur]

I forgot all about cache servers. I'm sure somebody already talked about Squid.

[@fiveangle]

In 2001 128MB was the default largest ram size that wasn't insane cost, and 512MB was like the largest, most premium sticks you'd have to sell your first born for. I recall our Dell developer configuration maxing out at 384MB of RAM support. In those days IIS was the web server top dog to beat and apache was an also-ran that hadn't really found it's space in the enterprise just yet.

[@UpLateGeek]

Hopefully you can get it working in transparent mode. We used to have Blue Coat proxies at my work in the mid-2000s, but I moved into the network department just as they were replacing them with Ironport proxies. Ours were always using explicit mode rather than transparent, so it would be interesting to see how that works. I remember looking into transparent mode, but it's been so long I can't remember much about it. In any case, ours were setup for outbound requests, not caching inbound requests, which it sounds like you want, hence transparent mode makes a lot more sense. The thing with the two boot options was just two partitions with the newer and older versions of system software. During a software upgrade, the older partition would be upgraded and the newer one kept as the backup. When it's finished upgrading, it sets the boot partition to the upgraded one and reboots, then if there's a problem, you can roll back by just manually booting from the original partition.

[@hayzeproductions7093]

6:00 proprietary os on a rom chip was too much at the time for web networking back in very early 2000's.. We worked with IDE hard drives back then as a standard (along with using jumper pins for the bios to detect master and slave drives). Having 1Gig of SD ram or DDR1 ram is far fetched for a system "Top of the Line". Specially for HTML back in early 2000's What we have now for HTML? A LOT with HTML5 including PHP 8.X. Having 1gb of ram for a basic webserver or cache server is pretty much not enough for a wordpress site now days.

[@TimmyHD]

I recently pulled a slot loading Pentium 3 processor out of one of our old prod PCs and was fascinated at how they work... We're still running Windows 98 in prod at one facility. Proprietary hardware/software combo makes it not feasible to update at the moment.

[@pikewerfer]

Never heard of those - looks cool

[@anancient24]

32:24 Casually throws devil's horns 🤘, and you're right; those Nortel switches are metal as f***! Alternatively you were signing 🤟the ASL letters I-L-Y for "I love you", which is also valild. Those switches are lovable!

[@tommajor2940]

Hey! It was great seeing you and "the serial port" in Chicago. I was the guy with the older Data Comms exhibit (Adtran T1 and TDM stuff with the Fireberd 6000A test sets). Any way it looks like this console screen menu (Cache server) will allow you to do some PCAP (Packet Capture) commands! I would bet that you could probably produce some packet capture files and then ftp them over to a computer that you can run Wireshark from. Wireshark would be my preferred method of showing the packet data. It might be interesting to see what the Jolt data looks like with a capture.

[@clabretro]

hi! was great seeing you as well. that's interesting about the PCAP commands, I'll have to try that

[@tommajor2940]

@@clabretro There might be some options to run Packet Capture from the Web interface and then just download the results to your local PC via web. Some firewalls would have this feature in order for the Admin to troubleshoot rulesets that may not be working as expected. If you've ever looked a PFSense it has that option to capture on any interface (Meaning LAN or WAN side).

[@RachaelSA]

At ISPs I worked at back in the day I used to do a similar thing with squid running in reverse, I could make acls and stuff, and wrote my own logger to log every request to SQL, and it could even protect the IIS machines when new exploits came out that didnt have patches yet.

[@_chrisr_]

The point of the cache was to allow a server that was serving content like asp or other dynamic content not to be distracted by requests for static content such as.html or .jpg etc. The transparent option will probably requre you to enable port mirroring on the switch for the port that the server is on -this allows the traffic destined for the server to be seen by the cache engine and the cache engine will respond on behalf of the actual server. I presume it blocks the TCP negotiation with the actual server to stop the actual server doing any meaningful work. I recall installing a network security device that did this - the device would inspect the outbound request and if it didn't like it would just respond as if it was the site that the user had attempted to access with a blocked message.

[@DanielTheRat]

Great video would love to see if 2 or 3 jolt2 instances could bring the ibm thinkcente to its knees since one brings it to 40% utilisation.

[@Ben333bacc]

Forgive me if someone already may have said it but: The reason you are seeing 40% ish from the DOS attack is because the TCP stack on XP isn't multi-threaded. If you disabled HT support, it'd use 100%. Same reason most video games back in the day, or any piece of non SMP software can't use 100%... Can only use 100% divided by number of cores + threads. So, 50% for a ht P4.

[@jeffmoss26]

Wow! We had one of these at my school district

[@arizonapalms]

1U x86 based appliance that does 1 specific task? Yep this is my jam The yellow face reminds of the old Symantec firewalls too lol

[@clabretro]

x86 appliance just can't be beat

[@revision386]

I know someone from a small local ISP and they had Google cache servers that had SCSI drives .

[@dannooo548]

Would it have been common in that era to have a site with a bunch of slow PHP code or something else server-side that would really show off the caching capability, versus a static HTML page? Maybe the goal is not crazy-high requests per second but caching the response from some big database query.

[@clabretro]

yes! I'll be covering a test with a DB-backed app like that in future cache videos

[@andersama2215]

On the cache invalidation thing, refreshing like you did around 35 minutes in doesn't really show us the cacheflow in action because (now I can't speak to that browser), but if it's caching the result, doing a refesh shouldn't send requests to the server. I remember specifically some browsers had a "hard refresh" button or option/shortcut and that was supposed to ignore the browser's cache, not sure if that's what you were doing.

[@nickwallette6201]

Interesting results. I don't know what the vendor assumed was the ideal use-case for this, but it probably helped a lot to offload static content on web servers that were just starting to move to dynamic HTML generation. Caching the HTML, images, CSS, and JS files would spare the server having to cache them itself, freeing up that memory to cache databases or whatever data sources it used to generate dynamic content. Obviously this doesn't show much benefit with a load-testing tool that is demonstrating the caching performance of apache vs the caching performance of the CacheFlow -- although it is impressive that a Socket 7 box is keeping up with that P-III server. It could also be very helpful serving as a local-side WAN accelerator, caching static content that might be accessed from multiple clients, and saving upstream bandwidth. I'm sure you would see a HUGE benefit if you put the web server on the far end of a T1 link. :-)

[@clownhands]

Awesome video! Would love to see you do a video on Netscape Enterprise Server / iPlanet software. They were the Rolls Royce of web server, ldap server, and mail server at the time.

[@clabretro]

yes, eventually!

[@kaitlyn__L]

Oh, I’d forgotten how websites around 2000-2002 would take a lot longer to respond for anything that wasn’t the homepage or the day’s new posts. You’d get the text after 10 seconds, then 30-60 seconds later the pictures would finally start loading in. I’d completely forgotten “waiting for server response”, haha 😅 I love how 5 concurrent users was a reasonably heavy (but not catastrophic) test for Networking magazine’s benchmark.

[@autarchprinceps]

It's a little weird to try to test the effectiveness of a HTTP proxy, by using a UDP load/DoS tool like jolt. Ain't gonna cache that. It might act as a block to those packages sure, but certainly not when used as a proxy only, and in general that seems more like a firewall thing to handle. I guess today a CDN would protect a system from that, but again not with the HTTP cache part of the service, but simply by not having a UDP port open.

[@jfbeam]

That's how people killed many a website back in the day. One lowly kid on a 28.8 dialup connection could run one tiny badly written program and kill a "major" website.

[@autarchprinceps]

@@jfbeam Well if you host your website on a Windows server, you deserve nothing else.

[@EpsilonsReviews]

When in doubt of disks, Clonezilla the heck outta them!