wow such numbers
on recruiters
From: Stunningly Inept Recruitor <robert#######@gmail.com>
To: Apparently Everyone I know on the Internet
Subject: Exciting pre-IPO Opportunity
Hi $username
I was wondering if you may be interested in an incredibly exciting IPO opportunity in SF. We are very near profitable. Here are some of the interesting numbers we deal with:
Sure, why not? Since this is a cold email and I know very little at this point about your business, software architecture, technical goals, debt, or operating constraints: budget, team, core skills, rate of growth, etc … I’m sure I have all the context needed to responsibly evaluate your metrics.
So, let’s get to it.
Our ad servers frquently respond in in under 4ms — they are required to respond in 160ms. Most of our work is done in realtime.
Oh, I see what you did there! I like how you’re simultaneously letting me know that you’re in advertising and that you care about performance. I’m trying to remember which percentile frequently falls into? Was that p5? p10? p50? Maybe you have some sort of distribution and deviation rates you’re looking to narrow? Also, mentioning realtime makes me wonder if that performance metric is really queue latency rather than query latency.
We carry around 200TB of used storage in our hadoop cluster. That’s 200 terabytes.
Who the hell are you targeting that you also feel the need to clarify that TB is terabytes? 200TB sounds impressive until you realize you can easily have 300+TB of SSDs in a single rack and even more if you’re happy with slow media.
We have around 250 machines.
Such Wow! If only I had a way to contextualize that number. Are they busy? 10%, 50%, 98%? DC or EC2? One DC/AZ? Multi-DC/AZ? Active-Passive or Multi-Master? Front-End? Back-End? Message-Queue? Storage? Faith-Based Consistency?
Our hadoop cluster is currently suffering because it’s not on 10GbE. We’re bottoming out our switches?
I’m just going to go ahead and infer a couple of things here:
- Okay, so you’re not in AWS, Azure, etc.
- You’re almost certainly in a single-DC, good luck with that IPO.
- You’re either in Rackspace (or similar) or you need to be hiring Infra/Ops people like there’s no tomorrow.
- Unless your switches are attached to ‘69 Impala with hydraulics, I highly doubt you’re bottoming out the switch. If you are in fact saturating your switches, cool story: you might want to re-evaluate your partitioning strategy in hadoop.
We devote slightly under a terabyte of ram to Redis.
Shit. I don’t even know where to start. I guess that seems like a lot to you? Is this a point of pride or is coming from a place terror: oh shit, one day we woke up with this giant Redis cluster and fork/persist is running a train on our p99+" ? How many of those 250 machines are running Redis? How many of those machines are only running ~1.25 of N cores hot?
PostgreSQL is too slow for us. (yes, really).
Cool story? This is one of those places you really need a link to a talk/blog from one of your engineers contextualizing this shit. In the meantime insert the obvious curious comments about partitioning strategy, fast-storage, etc.
It’s a challenge integrating DNS into our environment because resolving requirements are simply too fast for BIND (yes, really).
Wat? Similar to above — without an actual engineer providing context this just sounds like crazy talk.
